Vegetation structure significantly impacts the degree of anisotropy of the scattered radiation field. The proper analysis of multiangular data, such as those provided by the Multi-angle Imaging SpectroRadiometer (MISR) instrument on board Terra, could thus in principle yield statistical information on the structure of the observed environment. Preliminary investigation in this direction suggests that useful information on the heterogeneity of the vegetation can be retrieved at the subpixel scale.

The spatial resolution of remotely sensed imaging devices becomes higher and higher. Since the launch of QuickBird, during the year 2001, panchromatic images of 66 cm are acquired. As those improvements have been performed very quickly, methods for an automatic processing of those images are not yet available. The improvement of the spatial resolution enables the detection of new kind of objects. For example, instead of detecting forests, trees are. New applications, notably an accurate survey of the environment during exceptional events (flooding, fire, ...) are conceivable. However, the areas which used to be homogeneous within a 10-meter resolution, are then heterogeneous. Consequently, commonly used methods, such as classification for example, are less efficient. It is urging to propose techniques for an automatic exploitation of this kind of images. In this paper, we propose to add, before the commonly used processes a pre-processing to simplify an image by the diminution of the heterogeneity within regions corresponding to a unique entity, while keeping the borders. For so doing, we compare several filters, linear and non linear. In particular, we use a morphological pyramid based filtering. An example is shown on a QuickBird image acquired over Berne, and comparison of all the filters is done.

In this paper a denoising technique for multispectral images exploiting interband correlations is proposed. A redundant wavelet transform is applied and denoising is applied by thresholding wavelet coefficients. A scale adaptive threshold value is obtained by exploiting the interband correlation of the signal. First, the coefficients from different bands are multiplied. For these products, the signal and noise probability density functions (pdf) become more separated. The high signal correlation between bands is exploited further by summing these products over all bands, in this way separating noise and signal pdfs even more. The noise pdf of the proposed quantities is derived analytically and from this, a wavelet threshold is derived. The technique is demonstrated to outperform single band wavelet thresholding on multispectral remote sensing images.

In this paper, two different approaches for horizon picking are examined. The first one is a simple line detection algorithm applied to the full resolution image. The second algorithm is a multiscale line detection algorithm, based on the wavelet transform of the iamge. Both full resolution and multiresolution line detection algorithms are applied to 2D seismic images and compared in terms of their performance. Results show that the full resolution line detection approach outperforms the multiresolution approach.

Many image fusion algorithms such as Principal Component Analysis (PCA), Multiplicative Transform, Brovey Transform, IHS Transform and wavelet transform have been presented in order to fuse low-resolution multispectral data with high-resolution panchromatic data. More recently Yun Zhang has presented a new algorithm for the fusion of Landsat ETM and Ikonos data respectively. In this study we compare the efficiency of five of the above fusion techniques and more especially the efficiency of PCA, Multiplicative Transform, Brovey Transform, IHS Transform and Pansharp for the fusion of Landsat ETM data. The area of interest is situated in Western Peloponnese near the city of Pyrgos. The broader region combines at the same time the characteristics of an urban, a coastal and a rural area. A Landsat 7 ETM cloud free subscene taken in the morning of July 28, 1999, was used in this comparative study. For each fused image we have examined: a) the optical qualitative result, b) the statistical parameters of the histograms of the various frequency bands, especially the standard deviation c) the amplitude spectrums of the frequency bands. All the fusion techniques improve the resolution and the optical result but according to the statistical analysis the Brovey and the Multiplicative techniques do not improve the contained information in the fused images. The IHS fusion technique seems to have the best optical result, increases the sum of the contained information but provoke changes to the colors of the original RGB image. The PCA fusion technique seems better in discriminating between the coastal zone, the urban area and the rural area and maintains the natural colors. The Pansharp fusion technique gives the best results without changing at all the statistical parameters of the original multispectral image.

Digital Elevation Models (DEM) have become important tools in many remote sensing applications, such as classification, defense, Geographic Information Systems, etc. But they are complex products to generate and they are still pervaded with errors and artifacts due to the generation techniques themselves or atmospheric problems. Thus their qualification for a specified application is not guaranteed. It is well known nowadays that the evaluation of the quality of a DEM is a challenging task, due to the variety of requirements depending on the applications and on the end-user. It remains a major field of investigation, where scientists always look for new tools for the analysis of DEM. The use of multiresolution techniques is one possible answer to this research. In the past decades it has been shown that natural landscapes exhibit fractal behaviors. Consequently it seems rather obvious and relevant to use techniques based on fractals and more generally on multi resolution concepts as a tool for understanding the geological nature of terrain. Thanks to the emergence of the use of the wavelet theory, researchers get interested again in fractals modeling for geo information processing and understanding. In this article our aim is to present the various analysis that are possible to lead on DEM thanks to multi resolution methods, in particular wavelet filtering and fractal dimension estimation.

The aim of Super-Resolution techniques is to produce a high-resolution image from a sequence of shifted low-resolution images. These low-resolution images are generally taken from slightly different viewpoints, which can result in some new information from one image to another. Super-resolution is a very attractive research area and finds its applications in many domains. We present in this paper a new algorithm to achieve image Super-Resolution suitable for sequences of satellite images. The shift between images is estimated using a new measure based upon the sharpness property of edges in the reconstructed image. The reconstruction is performed using a new approach to super-resolution based on the orthogonal set of Walsh-functions. In this paper we shall derive an expression for the missing information contained in an image of 2^M+1 pixels but absent in the lower resolution image of 2^M pixels. From this we prove how this information can be retrieved from a second image containing 2^M pixels, shifted from the first by an arbitrary amount. We then derive analytic models for the distortion of edge information by this process and use the blurring in the reconstructed image to perform accurate registration between the two shifted images at low resolution. We present results from this method based upon test imagery where the shift is known a priori to evaluate performance.

In this work a pixel-level fusion technique for enhancing the visual interpretation of multispectral images is proposed. The technique takes into consideration the inherent high correlation of the RGB bands of natural color images, a fact strictly related to the color perception attributes of the human eye. The method provides dimensionality reduction in the multispectral vector space, while the resulting RGB color image tends to be perceptually optimal. The proposed method is compared with two other existing techniques.

In the real world of remote sensing, rarely does the extracted object precisely match a stored template of that object. A certain level of uncertainty must be permitted between the stored template and the extracted object. One solution to deal with this uncertainty is to evaluate measures between the object and template. Many measures have been introduced independently in literature for evaluating the statistical nature of the extracted objects such as a variety of shape and texture measures. The object is measured and compared to similar measures taken from the template. This paper suggests using measures extracted through a joint comparative process of template and object. It also suggests using multiple measures from the joint class of measures as opposed to using an individual measure to determine the sufficiency of the match. In particular, this paper demonstrates the value of using multiple comparative shape measures as opposed to one particular shape measure to achieve confidence in a match. The multiple-shape measure approach uses a matched filter measure, a Procrustes metric, a partial-direct hausdorf measure, and percent-pixels same measure. Each shape measure gives slightly different insight about the shape comparison which allows more confidence in the match. An experimental result is given that demonstrates the implementation and usefulness of the multiple comparative measure approach for recognizing objects from remotely sensed imagery.

Among the most important types of information contained in radar and optical images are the textural features. In real life images these features are partially masked by noise that is always present in registered data, basically multiplicative in radar images and additive in optical ones. Thus, one of the basic steps in processing remote sensing data is the filtering of the observed image. However, despite the fact that a lot of filters have been already designed, relatively little attention has been paid to texture preservation properties of noise attenuation methods. Thus, there are the following actual tasks: 1) to analyze the texture preservation properties of different filters; 2) to design image processing methods that are able to preserve texture features simultaneously with effective noise suppression. In this paper, the texture feature preserving characteristics of different filters are examined using a set of texture samples, different noise levels and a set of parame-ters including spatial correlation and higher order statistics. The traditional locally adaptive two-state hard switching filters are modified to the three-state ones where texture is considered as a particular class. For “detection” of texture regions, special, rather simple, classifiers that are based on joint analysis and processing of the two local activity indicators are proposed. The recommendations concerning the parameter setting of the classifiers are given. All this provides an appropriate trade-off of the designed filter properties. It improves the PSNR for entire image in comparison to the component filters used within the three-state local adaptation framework. Local PSNRs for the considered types of image fragments are practically the same or even better than for the filter type recommended for the processing of the corresponding classes. Real life image examples are presented to demonstrate the efficiency of the proposed filter.

Recently the lake area detection has been a popular topic for time series remote sensing images analysis. The two-dimensional Markov model is one of the efficient mathematical models to describe an image especially when the within-object interpixel correlation varies significantly from object to object. The unsupervised Region Growing is a powerful image segmentation method for use in shape classification and analysis. In this paper, the Region Growing method based on two-dimensional Autoregressive Moving Average (ARMA) model is proposed for lake region detections. Some of the statistical techniques, such as Gaussian distributed white noise error confidence interval, and sample statistics based on mean and variance properties have been used for thresholding during calculations. The linear regression analysis with least mean squares estimation is still of ongoing interest for statistical research and applications especially with the remote sensing images. The LANDSAT 5 database in the area of Italy's Lake Mulargias acquired in July 1996 was used for the computing experiments with satisfactory preliminary results.

The segmentation process of satellite imagery becomes currently a significant step in remote sensing with the arrival of very high spatial resolution images. Indeed, the arrival of these images enables a new capability to study a range of non-observable objects until now. Using high-resolution imagery should make it possible to detect man made features (such as buildings and roads) or detailed components of vegetation (such as trees or heterogeneous woodlands) in an easier way than conventional data. In this paper, we present a brief review of segmentation techniques, the principal advances in earth observation technology, and the evolution of the high-resolution technology. Also, we present a self-adapting method of segmentation of very high-resolution satellite images. This approach is based on a description of the image using graphs of adjacency and morphological processing to obtain suitable and significant computed components by the growth of regions. Finally we present some examples of the segmentation and the feature extraction done in some high-resolution images.

This paper describes a precise geometric correction method considering elevation effects for NOAA/AVHRR of GMS images, which is mandatory for long-term global environmental monitoring studies. First, using the so-called systematic geometric correction, the correspondences of sub-sampled image pixels to their map coordinates are calculated. And, the correspondences of sub-sampled map locations, which are the corner points of blocks, to image pixels are calculated to speed up the inverse transform to find for a pixel on the map coordinates to the corresponding pixel in the image coordinates using the bilinear interpolation of the four corner points of a block. For precise geometric correction, the residual errors of the systematic correction are measured using many GCP templates. GCP templates in the map coordinates are provide using DCW. Templates in the image coordinates are generated using the bilinear Interpolation. Also, the templates of high elevation areas are modified to include the elevation effects, using the height from GTOPO30 and satellite sensor geometry. Then, the residual errors are acquired by template matching and affine transform coefficients are calculated to remove the residual errors. And if the difference between the average error and each GCP is more than one pixel, these GCP’s are removed and new affine transform coefficients are recalculated iteratively until all errors reach within one pixel. Then, mapping of each pixel is done using the correspondence of four corner block points and image coordinates modified by affine transform, but for high elevation areas blocks are divided into pixels according to their elevation. The accuracy of within one pixel; i.e. 0.01 degree for NOAA/AVHRR and GMS/VIS and 0.04 degrees for GMS/IR is obtained for NOAA images received at Tokyo and the stitched ones received at Tokyo and Bangkok and also GMS full disk images.

The aim of investigation was developing the image registration algorithms dealing with the aerial and cosmic pictures taken in different seasons from differing view points, or formed by differing kinds of sensors (visible, IR, SAR). The task could not be solved using the traditional correlation based approaches, thus we chose the structural juxtaposition of the stable specific details of pictures as the general image matching technique. Structural matching was usually applied in the expert systems where the rather reliable results were based on the target specific algorithms, but our algorithms deal with the aerospace photographs of arbitrary contents for which the application specific approaches could not be used. The chosen form of structural descriptions should provide distinguishing between the similar simple elements in the huge multitudes of image contours, thus the descriptions were made hierarchical: we grouped the contour elements belonging to the separate compact image regions. The structural matching was carried out in two levels: matching the simple elements of every group in the first image with the ones of every group in the second image; matching the groups as the wholes. The top-down links were used to enhance the lower level matching using the higher level matching results.

A method for speckle reduction and segmentation of synthetic aperture radar (SAR) images is presented here. As a first step we consider a preclassification to a specific number of labels (classes). The second step (final classificatio) is the relaxation labeling process. The method can be considered as a fast unsupervised technique. We have worked in 3-look simulated and real ERS-1 amplitude images.

This paper discusses processing of interferometric signal to measure ocean topography. The method corrects for channel misregistration and geometric decorrelation using estimates of the current Earth geoid scenario. The channel misregistration is caused by slight change in the differential time delay between the signal received at one antenna with respect to the other. This algorithm corrects the misregistration over the entire swath with the Chirp-Z transform which resamples the signals appropriately. Another source of error, the geometric decorrelation (or baseline decorrelation), occurs because the targets within the resolution cell contribute different interferometric phases. In essence, the ground projected wavelengths are different for various look angles which produces a shift of the effective spectrum. This is corrected by shifting the spectra relative to one another and by applying filters to eliminate the non-overlapping part of the spectra. However, the co-registration and the spectral shift require the estimation of the current look and incidence angles. We use the Earth Ellipsoid WGS-84 and the Geoid EGM-96 to estimate the geometric parameters to describe the various viewing scenarios encountered around the Earth Geoid. We finally discuss the implications on the signal processing algorithm.

This paper presents an application of Fourier Descriptors and Fuzzy Logic for the recognition of archeological artifacts in Ground Penetrating Radar (GPR) images of a surveyed site. 2-D GPR survey images of a site are made available by NASA-SSC center. The buried artifacts in these images appear in the form of hyperbolas which are the results of radar backscatter from the artifacts. The Fourier Descriptors of an image are applied as inputs to a Fuzzy C-Mean Classifier (FCMC). The FCMC algorithm has to recognize different types of shapes, in order to separate hyperbola-like shapes from non-hyperbola shapes in the sub-surface images. The procedure consisted of removing background noise using a suitable threshold filter, locating the separate shapes in the image using N₈(p) connectivity algorithm, calculating a short sequence of Fourier Descriptors (FD) of each isolated shape, and obtaining an unsupervised classification by applying Fuzzy C-Mean clustering algorithm to the FD sequences. The classes obtained depend upon the requirements of the user, namely, two classes of hyperbola/no-hyperbola objects, or several classes from symmetric hyperbolas to total rejects could be obtained. The results consisting of recognized hyperbolas indicate the presence of buried artifacts. Also, our previous results of supervised FD-Neural Network (FD-NNC) published in the proceedings of SPIE 2002 are compared with unsupervised FD-FCMC. The compared results in terms of the quality of classification are presented in this work.

This paper presents a method for recognition of Noisy Subsurface Images using Morphological Associative Memories (MAM). MAM are type of associative memories that use a new kind of neural networks based in the algebra system known as semi-ring. The operations performed in this algebraic system are highly nonlinear providing additional strength when compared to other transformations. Morphological associative memories are a new kind of neural networks that provide a robust performance with noisy inputs. Two representations of morphological associative memories are used called M and W matrices. M associative memory provides a robust association with input patterns corrupted by dilative random noise, while the W associative matrix performs a robust recognition in patterns corrupted with erosive random noise. The robust performance of MAM is used in combination of the Fourier descriptors for the recognition of underground objects in Ground Penetrating Radar (GPR) images. Multiple 2-D GPR images of a site are made available by NASA-SSC center. The buried objects in these images appear in the form of hyperbolas which are the results of radar backscatter from the artifacts or objects. The Fourier descriptors of the prototype hyperbola-like and shapes from non-hyperbola shapes in the sub-surface images are used to make these shapes scale-, shift-, and rotation-invariant. Typical hyperbola-like and non-hyperbola shapes are used to calculate the morphological associative memories. The trained MAMs are used to process other noisy images to detect the presence of these underground objects. The outputs from the MAM using the noisy patterns may be equal to the training prototypes, providing a positive identification of the artifacts. The results are images with recognized hyperbolas which indicate the presence of buried artifacts. A model using MATLAB has been developed and results are presented.

Hyperspectral data can potentially provide greatly improved capability for discrimination between many land cover types, but new methods are required to process these data and extract the required information. Data sets are extremely large, and the data are not well distributed across these high dimensional spaces. The increased number and resolution of spectral bands, many of which are highly correlated, is problematic for supervised statistical classification techniques when the number of training samples is small relative to the dimension of the input vector. Selection of the most relevant subset of features is one means of mitigating these effects. A new algorithm based on the tabu search metaheuristic optimization technique was developed to perform subset feature selection and implemented within a binary hierarchical tree framework. Results obtained using the new approach were compared to those from a greedy common greedy selection technique and to a Fisher discriminant based feature extraction method, both of which were implemented in the same binary hierarchical tree classification scheme. The tabu search based method generally yielded higher classification accuracies with lower variability than these other methods in experiments using hyperspectral data acquired by the EO-1 Hyperion sensor over the Okavango Delta of Botswana.

A novel classification scheme for SAR images based on the perceptual classification of image patterns in the Discrete Hermite Transform domain has been developed. In order to obtain the DHT referred to a rotated coordinate system the set of coefficients of a given order are mapped through a unitary transformation based on the generalized binomial function. This representation allows a perceptual classification, including constant patterns (0-D), oriented structures (1-D), and non-oriented structures (2-D). Classification is based on light adaptation and contrast masking properties of the human vision. Finally, classification is improved by means of a probabilistic approach based on Markov Random Fields.

A general problem of supervised remotely sensed image classification assumes prior knowledge to be available for all thematic classes that are present in the considered data set. However, the training set representing this prior knowledge usually does not really describe all the land cover typologies in the image and the generation of a complete training data set would be a time-consuming, difficult and expensive task. This problem may play a relevant role in remote sensing data analysis, since it affects the classification performances of supervised classifiers, that erroneously assign each sample drawn from an unknown class to one of the known classes. In the present paper, a classification strategy is proposed, which allows the identification of samples of unknown classes, through the application of a suitable Bayesian decision rule. The proposed approach is based on support vector machines for the estimation of probability density functions and on a recursive procedure to generate prior probabilities estimates for both known and unkown classes. For experimental purposes, both a synthetic and a real data set are considered.

It has long been shown the effectiveness of artificial neural networks to solve highly non-linear problems such as land-cover classification based on multispectral imagery. However, due to the large amount of data that is processed within this kind of applications, it is desirable to design networks with the lowest number of neurons that are capable to separate all of the given classes. At present, there are several methods intended to determine this optimal network. Most of them involve adjoining or pruning hidden neurons followed by further training in iterative fashion, which is generally a very slow process. As an alternative, the approach described in this paper is based on the computation of centroids of relevant clusters for each class samples through the well known clustering method ISODATA. A proper tessellation of the ISODATA centroids allows first the determination of the minimum number of neurons in the first hidden layer that are required to effectively separate all of the classes; and secondly, to compute weight and bias parameters for such neurons. Then, the minimum network required to perform the logic function that combines the halfspaces generated by the first layer into class-discriminant surfaces is determined via a logic function reduction method. This approach is much faster than that of current methods because it allows to determine the optimum network size and compute weight and bias parameters without further iterative adjustments. The procedure was tested with landsat 7 Enhanced Thematic Mapper Plus (ETM+) data. Results indicated that (1) the network exhibits good generalization behavior and (2) classification accuracies do not depend on the class boundary complexity but only on the class overlapping extent.

The next-generation NOAA/NESDIS GOES-R hyperspectral sounder, now referred to as the HES (Hyperspectral Environmental Suite), will have hyperspectral resolution (over one thousand channels with spectral widths on the order of 0.5 wavenumber) and high spatial resolution (less than 10 km). Hyperspectral sounder data is a particular class of data requiring high accuracy for useful retrieval of atmospheric temperature and moisture profiles, surface characteristics, cloud properties, and trace gas information. Hence compression of these data sets is better to be lossless or near lossless. Given the large volume of three-dimensional hyperspectral sounder data that will be generated by the HES instrument, the use of robust data compression techniques will be beneficial to data transfer and archive. In this paper, we study lossless data compression for the HES using 3D integer wavelet transforms via the lifting schemes. The wavelet coefficients are processed with the 3D set partitioning in hierarchical trees (SPIHT) scheme followed by context-based arithmetic coding. SPIHT provides better coding efficiency than Shapiro's original embedded zerotree wavelet (EZW) algorithm. We extend the 3D SPIHT scheme to take on any size of 3D satellite data, each of whose dimensions need not be divisible by 2^N, where N is the levels of the wavelet decomposition being performed. The compression ratios of various kinds of wavelet transforms are presented along with a comparison with the JPEG2000 codec.

This paper investigates on the development of an advanced method for lossless compression of hyperspectral data to be implemented on board of a space platform. An adaptive Differential Pulse Code Modulation (DPCM) method, jointly exploiting spectral and spatial correlation and utilizing space-oriented context-based entropy coding, is taken as starting point. The algorithm considered utilizes a "classified" DPCM approach, in which predictors, taking into account the statistical properties of the data being compressed, are preliminarily calculated and then adaptively selected or combined. Two fuzzy clustering algorithms are tested with the aim of finding the best algorithm to be employed in the initialization phase, which is the core of the "classified" DPCM compression procedure, may be performed off line so as to unaffect the computational complexity of the online procedure running on board. The final method utilizes a standard CCSDS-Rice space encoder and represents a good tradeoff between compression capability and computational complexity. Overall coding performances, as well as differences between the two fuzzy clustering algorithms, are reported and discussed through extensive experiments carried out on four hyperspectral AVIRIS images.

High resolution images are nowadays a common source of data for many different applications; let us consider, for instance, hyperspectral images for remote sensing and geographic information systems. This kind of images allows for exhaustive analysis and provides good classification performance due to their high resolution (either bits per pixel, spatial, or spectral resolution). Nevertheless, this same high resolution, as well as their huge size, imposes a large demand of memory capability and channel bandwidth. To deal with this problem, lossy encoding of such images may be devised. Well known lossless and lossy image coding techniques have been used, but remote sensing and geographic information systems applications have some particular requirements that are not taken into account by the classical methods. There is therefore a need to investigate new approaches of image coding for these applications.

Mathematical morphology is a classic nonlinear image processing technique that has been successfully applied to analysis and classification of remotely sensed grayscale image data. The extension of basic morphological operations (i.e. erosion and dilation) to multi/hyperspectral imagery is not straightforward. In our approach, we treat the data at each pixel as a vector and impose a partial ordering of vectors in the selected vector space, based on their spectral purity. As a result, basic morphological operations are defined by extension, allowing joint spatial/spectral analysis of remotely sensed multispectral data. In this paper, we introduce the concept of scale-orientation morphological profile, and explore its application to mixed-pixel analysis and classification of hyperspectral data. The effectiveness of the proposed approach is assessed by using both simulated and real hyperspectral datasets collected by the NASA/Jet Propulsion Laboratory Airborne Visible-Infrared Imaging Spectrometer (AVIRIS). The proposed method is successfully applied for the purpose of land-cover classification and delineation of agricultural fields located at the Salinas Valley in California.

The high spectral and high spatial resolution, intrinsic to hyperspectral remote sensing, result in huge quantities of data, which slows down the data processing and can result in a poor performance of classifiers. To improve the classification performance, efficient feature extraction methods are needed. This paper introduces a set of features based on the discrete wavelet transform (DWT). Wavelet coefficients, wavelet energies and wavelet detail histogram features are employed as new features for classification. As a feature reduction procedure, we propose a sequential floating search method. Selection is performed using a cost function based on the estimated probability of error, using the Fisher criterion. This procedure selects the best combination of features. To demonstrate the proposed wavelet features and selection procedure, we apply it to vegetation stress detection. For this application, it is shown that wavelet coefficients outperform spectral reflectance and that the proposed selection procedure outperforms combining the best single features.

One of the most challenging task underlying many hyperspectral imagery applications is the spectral unmixing, which decomposes a mixed pixel into a collection of reflectance spectra, called endmember signatures, and their corresponding fractional abundances. Independent Component Analysis (ICA) have recently been proposed as a tool to unmix hyperspectral data. The basic goal of ICA is to find a linear transformation to recover independent sources (abundance fractions) given only sensor observations that are unknown linear mixtures of the unobserved independent sources. In hyperspectral imagery the sum of abundance fractions associated to each pixel is constant due to physical constraints in the data acquisition process. Thus, sources cannot be independent. This paper address hyperspectral data source dependence and its impact on ICA performance. The study consider simulated and real data. In simulated scenarios hyperspectral observations are described by a generative model that takes into account the degradation mechanisms normally found in hyperspectral applications. We conclude that ICA does not unmix correctly all sources. This conclusion is based on the a study of the mutual information. Nevertheless, some sources might be well separated mainly if the number of sources is large and the signal-to-noise ratio (SNR) is high.

We test a procedure for endmember extraction on a synthetic hyperspectral image. The procedure uses the Autoassociative Morphological Memories (AMM) as detectors of morphological independence conditions. To validate it we apply Convex Cone Analysis (CCA) to the same data. To generate the validation data, we synthesize the ground truth abundance images as the simulation of gaussian random fields and we use as ground truth endmembers some reflectance spectra obtained from the USGS repository.

This paper discusses the potential of radar signal to characterize the bare surface roughness in arid or semi-arid regions. Two different scales have been considered with two microwave sensors: high resolution ERS/SAR and low resolution ERS Wind Scatterometer instruments. Ground truth measurements were acquired over different arid sites in the South of Tunisia. An empirical approach is proposed to derive the surface roughness from SAR measurements. In this approach, the surface roughness is characterized by a parameter called Zs. Then, a good correlation between SAR and WSC data is demonstrated. Using these two sensors, we are able to derive the backscattering signal versus incidence angle, in the cases of big sand dunes, rocky relief and others. An empirical model is then proposed to retrieve the sand dune percentage within the different cells of the WSC.

The paper presents a multisensor image fusion algorithm which extends the solutions proposed for pan-sharpening of multispectral (MS) data through intensity modulation, to the integration of SAR and multispectral imagery. The algorithm is based on the computation of the ratio between a speckle-filtered SAR image and a low-pass approximation, obtained by 'a-trous' wavelet decomposition, of the same filtered SAR image. This ratio modulates the intensity of the multispectral image, which is obtained by applying a linear transformation, i.e., a generalized IHS transform, to the original MS data. The modulated intensity image substitutes the original intensity image of the multispectral data and the inverse transform is applied to obtain the fused multispectral image. Experimental results are presented on Landsat ETM+ and ERS SAR images of an urban area. The results prove accurate spectral preservation on vegetated regions, bare soil, and also on textured areas (buildings and road network) where information from SAR enhances the fused result which can be successfully applied both for display and classification purposes.

We present a method for the simultaneous deconvolution and interpolation of remote sensing data in a single joint inverse problem. Joint inversion allows sparsely sampled data to improve deconvolution results and, conversely, allows large-scale blurred data to improve the interpolation of sampled data. Geostatistical interpolation and geostatistically damped deconvolution are special cases such a joint inverse problem. Our method is posed in the Bayesian framework and requires the definition of likelihood functions for each data set involved, as well as a prior model of the parameter field of interest. The solution of such a problem is the posterior probability distribution. We present an algorithm for finding the maximum of this distribution. The particular application we apply our algorithm to is the fusion of digital elevation model and global positioning system data sets. The former data is a larger scale blurred image of topography, while the latter represent point samples of the same field. A synthetic data set is constructed to first show the performance of the method. Real data is then inverted.

In this work, a new strategy for the analysis of hyperspectral image data is described and assessed. Firstly, the image is segmented into areas based on a spatial homogeneity criterion of pixel spectra. Then, a reduced data set (RDS) is produced by applying the projection pursuit (PP) algorithm to each of the segments in which the original hyperspectral image has been partitioned. Few significant spectral pixels are extracted from each segment. This operation allows the size of the data set to be dramatically reduced; nevertheless, most of the spectral information relative to the whole image is retained by RDS. In fact, RDS constitutes a good approximation of the most representative elements that would be found for the whole image, as the spectral features of RDS are very similar to the features of the original hyperspectral data. Therefore, the elements of a basis, either orthogonal or nonorthogonal, that best represents RDS, are searched for. Algorithms that can be used for this task are principal component analysis (PCA), independent component analysis (ICA), PP, or matching pursuit (MP). Once the basis has been calculated from RDS, the whole hyperspectral data set is decomposed on such a basis to yield a sequence of components, or features, whose (statistical) significance decreases with the index. Hence, minor components may be discarded without compromising the results of application tasks. Experiments carried out on AVIRIS data, whose ground truth was available, show that PCA based on RDS, even if suboptimal in the MMSE sense with respect to standard PCA, increases the separability of thematic classes, which is favored when pixel vectors in the transformed domain are homogeneously spread around their class centers.

This paper addresses the problem of the classification of hyperspectral remote-sensing images by means of Support Vector Machines (SVMs). In a first step, we propose a theoretical and experimental analysis that aims at assessing the properties of SVM classifiers in hyperdimensional feature spaces which are compared with those of other nonparametric classifiers. In a second step, we face the multiclass problem involved by SVM classifiers when applied to hyperspectral data. In particular, four different multiclass strategies are analyzed and compared: the one-against-all, the one-against-one and two hierarchical tree-based strategies. The experimental analysis has been carried out by using hyperspectral images acquired by the AVIRIS sensor on the Indian Pine area. Different performance indicators have been used to support our experimental studies, i.e., the classification accuracy, the computational time, the stability to parameter setting, and the complexity of the multiclass architecture adopted. The obtained results confirm the effectiveness of SVMs in hyperspectral data classification with respect to conventional classifiers.

In this paper, we propose a kernel-based approach for hyperspectral knowledge discovery, which is defined as a process that involves three steps: pre-processing, modeling and analysis of the classifier. Firstly, we select the most representative bands analyzing the surrogate and main splits of a Classification And Regression Trees (CART) approach. This yields three datasets with different reduced input dimensionality (6, 3 and 2 bands, respectively) along with the original one (128 bands). Secondly, we develop several crop cover classifiers for each of them. We use Support Vector Machines (SVM) and analyze its performance in terms of efficiency and robustness, as compared to multilayer perceptrons (MLP) and radial basis functions (RBF) neural networks. Suitability to real-time working conditions, whenever a preprocessing stage is not possible, is evaluated by considering models with and without the CART-based feature selection stage. Finally, we analyze the support vectors distribution in the input space and through Principal Component Analysis (PCA) in order to gain knowledge about the problem. Several conclusions are drawn: (1) SVM yield better outcomes than neural networks; (2) training neural models is unfeasible when working with high dimensional spaces; (3) SVM perform similarly in the four classification scenarios, which indicates that noisy bands are successfully detected and (4) relevant bands for the classification are identified.

In this paper, we propose a new approach to the classification of hyperspectral images. The main problem with supervised methods is that the learning process heavily depends on the quality of the training data set. In remote sensing, the training set is useful only for simultaneous images or for images with the same classes taken under the same conditions; and, even worse, the training set is frequently not available. On the other hand, unsupervised methods are not sensitive to the number of labelled samples since they work on the whole image. Nevertheless, relationship between clusters and classes is not ensured. In this context, we propose a combined strategy of supervised and unsupervised learning methods that avoids these drawbacks and automates the classification process. The method is based on the general formulation of the expectation-maximization (EM) algorithm. This method is applied to crop cover recognition of six hyperspectral images from the same area acquired with HyMap spectrometer during the DAISEX99 campaign. For classification purposes, six different classes are considered in this area: corn, wheat, sugar beet, barley, alfalfa, and soil. Classification accuracy results are compared to common methods: ISODATA, Learning Vector Quantization, Gaussian Maximum Likelihood, Expectation-Maximization, and Neural Networks. The good performance confirms the validity of the proposed approach in terms of accuracy and robustness.

We present a modification to the Spectral Angle Mapper (SAM) to perform mixed pixel target detection in hyperspectral imagery. Our method uses the linear mixing model to define a new set of coordinates using endmember spectra. We show that spectral angles measured in this new coordinate system differ from those measured in full band space, and that there is a natural interpretation of angles involving mixed pixels in these new coordinates. We include experimental evidence that our method outperforms traditional SAM in cases involving both subpixel and shaded target situations.

When target motion is confined to a two-dimensional plane during coherent processing intervals, the adaptive joint time-frequency algorithm is shown to be an effective method for achieving rotational motion compensation in ISAR imaging. We illustrate the algorithm using both simulated and measured experimental radar data sets. The results show that the adaptive joint time-frequency algorithm performed very well in achieving a focused image of the target. Results also demonstrate that adaptive joint time-frequency techniques can significantly improve the distorted ISAR image over what can be achieved by conventional Fourier transform methods when the rotational motion of the target is confined to a two-dimensional plane. This study also adds insight into the distortion mechanisms that affect the ISAR images of a target in motion.

A new unsupervised texture classification method based on the genetic algorithms (GA) is proposed. In the method, the GA are employed to determine location and size of the typical textures in the target image. The proposed method consists of the following procedures: (1) the determination of the number of classification category; (2) each chromosome used in the GA consists of coordinates of center pixel of each training area candidate and those size; (3) 50 chromosomes are generated using random number; (4) fitness of each chromosome is calculated; the fitness is the product of the Classification Reliability in the Mixed Texture Cases (CRMTC) and the Stability of NZMV against Scanning Field of View Size (SNSFS); (5) in the selection operation in the GA, the elite preservation strategy is employed; (6) in the crossover operation, multi point crossover is employed and two parent chromosomes are selected by the roulette strategy; (7) in mutation operation, the locuses where the bit inverting occurs are decided by a mutation rate; (8) go to the procedure 4. Some experiments are conducted to evaluate classification capability of the proposed method by using images from Brodatz's photo album and actual airborne multispectral scanner. The experimental results show that the proposed method can select appropriate texture samples and can provide reasonable classification results.

Automatic information extraction requires a processing system to encapsulate the content of the image. This is a non-trivial task, because of the complexity of the information stored in images. In this paper satellite image enhancement and smoothing towards automatic feature extraction is accomplished through an effective serial application of anisotropic diffusion processing and alternating sequential filtering. Nonlinear diffusion processes can be found in many recent methods for image processing and computer vision. A robust anisotropic diffusion filtering is used with Tukey's biweight robust error norm for "edge-stopping" function, which preserves sharper boundaries than previous formulations and improves the automatic stopping of the diffusion. A well-known class of morphological filters, alternating sequential filtering is applied afterwards for a more extended enhancement and smoothing. The effective processing scheme is demonstrated with examples; Results appear promising.

The determination of spectral responsivities plays a significant role in analyzing and predicting the performance of digital imaging systems for remote sensing. For example, given the spectral response functions, we can readily obtain the colorimetric data from a camera corresponding to the remote illuminated objects. In this paper, we develop a filter-based optical system to estimate these functions. The design objective of this system is to effectively select a limited amount of spectral (or broadband) filters to characterize the spectral features of color imaging processes, which are contaminated with noise, so that the spectral response functions can be estimated with satisfactory accuracy. In our approach, a theoretical study is first presented to pave the way for this work, and then we propose a filter selection method based on the technique of orthogonal-triangular (QR) decomposition with column pivoting, called QRCP method. This method involves QR computations and a column permutation process, which determines a permutation matrix to conduct the subset (or filter) selection. Experimental results reveal that the proposed technique is truly consistent with the theoretical study on filter selections. As expected, the optical system with the filters selected from the QRCP method is much less sensitive to noise than those with other spectral filters from different selections. It turns out that our approach is an effective way to implement the optical system for estimating spectral responsivities of digital imaging systems.

The interpretation of mixed pixels is a key factor in the analysis of hyperspectral imagery. A commonly used approach to mixed pixel classification has been linear spectral unmixing. However, the question of whether linear or nonlinear processes dominate spectral signatures of mixed pixels is still an unresolved matter. In this paper we describe new methodologies for inferring land cover fractions within hyperspectral scenes, using nonlinear mixture modeling techniques based on support vector machines and neural network-based techniques. A comparative analysis of these mixture estimation methods to the standard linear mixture model has been carried out using a database of laboratory simulated-forest scenes. For the simulations, canopies of both opaque and translucent trees were simulated using objects mounted on stems. Two tree densities (sparse and dense) and three background colors (dark, white and green) were considered. Hyperspectral images of these simulated scenes were acquired by the Compact Airborne Spectrographic Imager (CASI), and the areal fractions of the main constituents calculated by the SPRINT canopy model were used for comparison. Our quantitative and comparative analysis reveals that nonlinear approaches outperform linear mixture model-based approaches, particularly in the scenes with translucent trees. As a result, this investigation suggests that nonlinear mixture models are needed to account for the multiple scattering between tree crowns and background for the laboratory simulated-forest scenes used in this study.

The extraction of man-made objects from remotely sensed imagery is a common application in remote sensing. Building detection is useful in territorial planning, mapping and Geographic Information Systems. Nevertheless these features are difficult to recognise in satellite data because of their variations in structure and size and especially because of the spatial resolution of the imagery. IRS panchromatic data, with 5,8 meters pixel size, was the higher spatial resolution sensor in civil applications until the Ikonos imageries distribution. Several approaches have been proposed for building detection in aerial images. Buildings cast a shadow in some direction and that is why many authors have employed shadows to detect constructions. Other authors use shadows to verify them, once they have been detected by some other techniques. This work focus on shadows detection probabilistic methods: it is found that digital supervised classification of the first principal component obtained from the application of a principal component analysis on the four channels of Ikonos allows identifying shadows and distinguishing them from other covers in the image. It is a fast and effective method and it can be implemented through tools available in commercial remote sensing software. This shadow detection system will provide cost-effectiveness in the inventorying of buildings, especially in areas of dispersed settlement, given that it significantly reduces fieldwork, and even can function as a support and test of the methods of automatic extraction of buildings from satellite images developed up to now.

To have an appropriate data base for the development of non-cooperative target identification techniques, airborne measurements were conducted with the mmW-SAR-system MEMPHIS over agricultural terrain with a variety of different fields and canopies of trees. Four different depression angles were used, ranging between 15° and 38°, which allows to determine important clutter parameters as a function of depression angle. During the measurements evaluated here, the transmit polarization was switched from pulse to pulse between horizontal and vertical. Most important for a comparison between different passes is a careful polarimetric calibration. This was done using a statistical method. Absolute amplitude calibration was achieved by means of trihedral corner reflectors. The SAR processing provides the user with several degrees of freedom. Apart from different cross range (Doppler) resolution cell sizes it is possible to create either single look or multi-look images and study the influence of averaging on reflectivity statistics. The results are valuable not only for discrimination and ATR algorithms but also for the development of polarimetric target/clutter simulation models. The paper describes the experimental set-up and discusses the evaluation methods. Typical results are presented and the implications on ATR methods are highlighted.

Infrared (IR) image analysis is important for extracting IR anomaly features in the fundamental experiments of Remote Sensing Rock Mechanics (RSRM) and for identifying omen for tectonic earthquake based on satellite remote sensing monitoring. Transferring the IR image into a raster contour map is very useful, but usually a raster contour map transferred is not smooth enough for the disturbance of noise and the limitation of spatial resolution of the image. A Gauss filter, G(i,j) ≡ ke i²+j²/σ², is selected for generating raster contour map. The practical appplication cases in image analysis and feature extraction for RSRM experiments suggest that the raster contour map is smooth enough for quantitative analysis, and is helpful for IR feature extraction and for anomaly identification. As compared to mean filter and median filter, the Gauss filter is effective both in filtering speed and in contour map's quality for the condition that filter width be 9~15 and σ be 0.2~0.6.

Exploiting a new distributed cooperative processing scheme where multiple processors cooperate in finding a global minimum, we have developed a new efficient maximum likelihood (ML) based calculation method for multitarget motion analysis under a fixed networked multisensor environment. The Track estimation of targets from sensor is a crucial issue in active dynamic scene understanding. Multitarget motion analysis, where there are multiple moving targets and multiple fixed sensors which only measure bearings of the targets, is to associate targets and sensor data, and estimate target tracks based on that association. This is NP-hard problem to obtain the optimal solution, as the method easily gets trapped in one of local optima. We applied the decentralized cooperative search technique to this problem, and proved our method effective. The method uses more than one processor, each of which has its own partial search space, searching multiple possibilities in parallel. This paper shows the current status of our research, and presents two prototypes of cooperative multi-agent systems for extended multi-target motion analysis.

In this paper we evaluate the performance of several image watermarking schemes applied to hyperspectral imaging. An image watermarking scheme based on JPEG2000 which can be also used to store and manipulate hyperspectral images is also described. Different watermarking schemes are tested in order to determine the suitability of each one for a specific hyperspectral image environment. The impact of classical GIS operations (namely zooming, cropping and compression) on the performance of each watermarking scheme is measured in terms of capacity and robustness. In order to do so, we study several possibilities for watermarking hyperspectral images, as all hyperspectral image bands should be taken into account. We also study the impact of watermarking in image quality, measured as usual by PSNR, but also by the degradation of classification performance. Compression, classification and watermarking are closely related to each other as decisions taken in one subject have a large impact on the others. Our results show that the newcomer JPEG2000 standard is a useful tool for both hyperspectral imaging and copyright protection purposes. The proposed watermarking scheme, which takes advantage of JPEG2000 standard capabilities, can be considered to be robust under the constraints defined by the integration of hyperspectral imaging with geographical information systems. JPEG2000 extensions defined by the standard related to this work are also considered.

In this paper, we propose a method able to fuse spectral information with spatial contextual information in order to solve “operationally” classification problem. The salient aspect of the method is the integration of heterogeneous data within a Multi-Layer Perceptron model. Spatial and spectral relationships are not explicitly formalized in an attempt to limit design and computational complexity; raw data are instead presented directly as input to the neural network classifier. The method in particular addresses new open problems in processing hyperspectral and high resolution data finding solution for multisource analysis. Experimental results in real domain show this fusing approach is able to produce accurate classification. The method in fact is able to handle the problem of a volumetric mixture typical of natural forest ecosystems identifying the different surfaces present under the tree canopy. The understory map, produced by the neural classification method, was used as input to the inversion of radiative transfer models that show a significant increase in the retrieval of important biophysical vegetation parameter.

The commonly presented speckle in Synthetic Aperture Radar (SAR) imagery makes it very difficult to be visually interpreted or directly used for quantitative application. Therefore, speckle reduction is a prerequisite for many SAR image processing and application projects. In this paper, we discuss a non-linear adaptive algorithm in detail. The key point of the algorithm is to sort the pixel values within a local window, to compute the sequences of standard deviations and means, and then to take the normalized differences between two successive standard deviation/mean ratios. Speckle detection is achieved by thresholding these differences. Speckle noise suppression is achieved by replacing the pixel’s digital value with rank-ordered local statistics. Experimental results shows that the proposed method demonstrates promising performance in comparison to some conventional SAR speckle filters in terms of the edge preservation, mean preservation, variance reduction and the visual effect.

Since 1999, very high spatial resolution data represent the surface of the earth with more details. However, information extraction by computer-assisted classification techniques proves to be very complex owing to the internal variability increase in land-cover units and to the weakness of spectral resolution. The increase in variability decreases the statistical separability of land-cover classes in the spectral space. Per pixel multispectral classification techniques are then insufficient for an extraction of complex categories and spectrally heterogeneous land-cover, like urban areas. Per region classification was proposed as an alternative approach. The first step of this approach is the segmentation. A large variety of segmentation algorithms were developed these last 20 years and a comparison of their implementation on very high spatial resolution images is necessary. For this study, four algorithms from the two main groups of segmentation algorithms (boundary-based and region-based algorithms) were selected. In order to compare the algorithms, an evaluation of each algorithm was carried out with empirical discrepancy evaluation methods. This evaluation is carried out with a visual segmentation of IKONOS panchromatic images.

Radar imaging provides an important advantage for the earth change observation independently of weather conditions. However, the recognition of some features as roads is more difficult leading thus to an ambiguous interpretation of the scene. In order to compensate the lack of features, the high spatial resolution panchromatic image is often used as a complementary data for improvng the quality of the radar image. The basic idea consists to extract features (details) from the panchromatic image by means of the High Pass Filter (HPF) in order to incorporate into the radar image. However, the difficult choice of the size and shape of the filter does not allow enhancing significantly the lack of features into the radar image. To resolve this problem, we propose the use of the a` trous algorithm as an alternative approach for extracting features from the panchromatic image. Its advantage lies to the local characterization of features by the wavelet coefficients. The radar-panchromatic composite image produced from the a` trous algorithm gives a better detection of lines, edges and field boundaries compared to the HPF method.

We propose in this paper the investigation of the change detection approaches based on the pixel level and the object level. The pixel level approach is based on the simultaneous analysis of multitemporal data, while the object level approach uses a comparative analysis of independently produced classifications of data. Thereby, the comparison is established by using the multilayer neural network classifier. Usually, the backpropagation algorithm is used as a training rule. In this paper, we investigate the use of the Kalman filtering (KF) as the training algorithm for detecting changes in remotely sensed imagery. By using SPOT images and evaluation criteria, the detailed comparison indicates that the KF algorithm is preferable compared to the BP algorithm in terms of convergence rate, stability and change detection accuracy.

The Principal Component Analysis (PCA) has been successfully applied to the characterization of noise in a sequence of frames and to the identification of bad pixels in an imaging array. Remote sensing scenery includes visualization through atmospheric turbulence and sea surfaces. These conditions produce spatial-temporal patterns that can be properly treated with the PCA method. A faint or weak source may be masked by the spatial features of the scene, or even by a fluctuating structure embedded on it. The PCA method is able to filter out these contributions related with global correlations of the set of data. In this sense the identification of the sources with the PCA is not based on the values of the Signal-to-Noise Ratio (SNR). It pays more attention to the spatio-temporal structure of the signals. Therefore, it is possible to identify sources below the classical SNR threshold. Another advantage is that the method corresponds with linear transformations, therefore it is easily implemented requiring a low computational effort. The approach used in this contribution is based on the same reasoning applied to the identification and classification of bad pixels. When the source is a point source, its image will fall on a small cluster of pixels (in the limit it will be only one pixel). This cluster is identified because the spatial-temporal evolution is different from the rest of the array. The method is applied to simulated sceneries as those found in images through atmospheric turbulence and detection of targets in sea images.

The main purpose of this work is to develop a new technique for image fusion based on the optimization of the Linear Mixture Model (LMM) through the algorithm known as Simulated Annealing (SA). The final result given by the algorithm is a fused image (FI) distinguished by a high spatial and spectral resolution. The algorithm proposed is being evaluated for the multispectral images registered by the Landsat 7 ETM+ sensor. In this study, the high spatial resolution image (HSRI) corresponds to the panchromatic image of this sensor, with a spatial resolution of 15m and the low spatial resolution image (LSRI) corresponds to the spectral bands TM1, TM2, TM3, TM4, TM5 and TM7, with a spatial resolution of 30m. As a result, it has been obtained images with a spectral resolution of the 6 bands and a spatial resolution of 15 m. The improvement in the quality of the fused images has allowed the identification of new, more homogeneous spectral classes.