This PDF file contains the front matter associated with SPIE Proceedings Volume 6787, including the Title Page, Copyright information, Table of Contents, Introduction, and the Conference Committee listing.

In this paper, a self-adaptive evolutionary clustering algorithm is presented. This algorithm uses the evolutionary programming (EP) to search the optimal clustering and bases on the principles of the K-means algorithm. The proposed self-adaptive evolutionary (SAEP) clustering algorithm self-adapts the vector of the step size appropriate for each parent. This is different from other genetic-based algorithms. The algorithm can minimize the degeneracy in the evolutionary process. The experimental results show that the KSAE clustering algorithm is efficient in the unsupervised classification of the multispectral remote sensing image.

In this paper, the problem of the detection of road networks in Synthetic Aperture Radar (SAR) images is addressed. Most of the previous methods extract the road by detecting lines and network reconstruction. Traditional algorithms such as MRFs, GA, Level Set, used in the progress of reconstruction are iterative. The tensor voting methodology we proposed is non-iterative, and non-sensitive to initialization. Furthermore, the only free parameter is the size of the neighborhood, related to the scale. The algorithm we present is verified to be effective when it's applied to the road extraction using the real Radarsat Image.

Through analyzing on data acquisition mechanism of LiDAR's multi-return signal, we deduce the relational model of microtopography and corresponding return signal, further research the rapid extract algorithms about terrain slopes, roughness and others terrain parameters. And then validate them by consecutive return signal data of GLAS, and field experiment.

Ideally, one objective of image fusion in remote sensing is to obtain high-resolution multispectral images with simultaneously the spectral characteristic of multispectral images and an enhanced spatial resolution. To date, numerous image fusion techniques have been developed. However, many methods may introduce spectral distortion, appearing as a change in colors between compositions of resampled and fused multispectral bands. To tackle this problem, some methods have taken the radiometric characteristics of sensors into account. This paper is an attempt to fuse high-resolution panchromatic and low-resolution mutlitspectral bands of the EO-1 ALI sensor. Starting from the analysis of spectral difference between ALI and other sensors, the authors present two methods which take into account the physical spectrum response of sensors during the fusion process: one is an improved fast intensity-hue-saturation (IHS) method with spectral adjustment according to sensor spectral response, and the other directly introduces sensor spectral response into the general component substitution image fusion method. An experiment based on ALI images has been carried out to demonstrate the effectiveness of the proposed approach. The fused images processed through the proposed methods have almost the same spatial resolution as panchromatic images and keep good spectral characteristics.

Atmospheric correction, which can retrieve water-leaving radiance, is an important preprocess in monitoring water quality from remote sensing data. The atmospheric correction algorithms developed by Gordon (1993, 1994) assume that water-leaving radiance of ocean waters in near infrared is zero. However, such an assumption is not applicable to inland waters, and usually leads to failure in atmospheric correction of remote sensing data of inland waters. Some scientists, based on some other assumptions, have developed some improved atmospheric correction algorithms which can be applied to coastal and inland waters. However, these algorithms can only get good results in specific areas. In order to get good results of atmospheric correction of remote sensing data of inland waters in China, an improved atmospheric correction algorithm is developed in this paper. This improved atmospheric correction algorithm assumes that water-leaving radiance in short-wave infrared is zero, which is based on the analysis of absorption and scattering characteristics of inland waters. This atmospheric correction algorithm is validated to have high applicable potentials by applied to concurrent MODIS data and in-situ measured reflectance spectra in Guanting Reservoir in North China.

Most of the feature extraction methods is complex and is easily influenced by some uncertain factors. Based on the idea of multi-resolution decomposing theory, a new statistical feature extraction algorithm based on gray multi-channel decomposition is proposed in the paper. The image's gray is decomposed into some channels. The gray of an image is partitioned into some gray regions, which is also called gray channels. The gray features on each channel are analyzed and extracted. Then all the gray features form feature vector. The multi-channel feature extraction method based on multi-resolution decomposition using wavelet theory is introduced at first. Each channel's gray feature is extracted and formed the feature vector. Targets can be well distinguished. But the method not has invariant property when image rotates. Referenced the thought of multi-resolution decomposition, a new multi-channel feature extraction based on gray is proposed in the paper. The four features, which is average of gray, variance, number of pixel and peakedness, are extracted and form the feature vector. The vector is used to identify targets. With the new method, the shortage of the former method is overcome. And the method not only has invariant property of rotation, but also has invariant properties of proportion and translation. The robustness is perfect. And the calculation is simple. The features of some infrared images of tank and helicopter are extracted with the new method. The results show its effectivity. It's helpful for image feature extraction and target recognition.

This paper presents a hardware-based implementation, which is specially designed for fast processing data cube of the novel imaging spectrometer, the Computed-Tomography Imaging Interferometer (CTII). CTII is the combination of the conventional Fourier Transformation Imaging Spectrometer (FTIS) and the Computed-Tomography Imaging Spectrometer (CTIS). CTII retains the advantages of both FTIS and CTIS such as high etendue, high spectral resolution. The price paid for above advantages is much enormous amounts of calculation compared with FTIS and CTIS. The CTII algorithm mainly includes two processing parts. One is Fast Fourier Transformation (FFT) Algorithm, the other is the Convolution Back-Projection (CBP) Algorithm. Since these two parts can be dealt separately, we introduced the pipeline processing technology. In this paper, we develop a hardware-based version of the CTII algorithm to improve its computational performance by resorting to Field Programmable Gate Arrays (FPGA) reconfigurable boards. The proposed FPGA design represents our first step toward the development of a system for onboard analysis of CTII hyperspectral imagery. The results we obtained demonstrate that the concept of the Computed-Tomography Imaging Interferometer (CTII) is correct, and the hardware-based version of the CTII algorithm is high efficiency in processing CTII data cube.

In this paper the simulated space-based high spectral resolution atmospheric infrared sounder (AIRS) infrared radiances with different cloud top heights and effective cloud fractions are used to demonstrate the measurement sensitivity and atmospheric profile retrieval performance. The simulated cloudy retrieval of atmospheric temperature and moisture derived from the statistical eigenvector regression algorithm are analyzed with different effective cloud fractions and different cloud height. The temperature and humidity root-mean-square error with cloud fraction ranging from 0.1 to 1.0 (with interval of 0.1) for cloud height (200, 300, 500, 700 and 850 hPa) known perfectly and cloud height error of 50 hPa are computed. Results show that the root-mean-square error of retrieved temperature and the mixed ratio of water vapor below the cloud top increase with effective cloud fraction. The retrieval accuracy of the cloud height error of 50 hPa decrease comparing with the cloud height known perfectly, while the temperature retrieval is more sensitive to cloud height error than humidity retrieval.

Aiming at the requirement of high frame frequency, real-time processing of detecting and tracking target in modern IR defense and detecting system, this article introduces the design and engineering implementation of a portable IR target automatic detecting and tracking system. Firstly analyzes the overall structure and introduce the composition and function in detail. Secondly puts forward the flow of low-altitude infrared small target detecting and tracking algorithm, then researches the background suppressing and detecting technology of far-distance weak small targets with low SCR. Finally aiming at the key technology of real-time processing with the limit of volume and power, a project of real-time signal processor that is reconfigurable based on DSP+FPGA is proposed and implemented, which sends the images and detecting results to host computer by USB2.0 interface. Besides, the hardware architecture, the target detecting flow in DSP, the driver and the firmware of USB2.0 are introduced too. The experimental result shows that this system can detect and track the target automatically and accurately, which satisfy the requirement of the real-time performance of the system.

Phase unwrapping is the key step in deriving topographic information from Interferometric SAR. We propose a region-cutting InSAR phase unwrapping method based on Shortest Distance First Principle(SDFP), the interferogram is cut into the residue regions which contain all the residues and the non-residue regions which contain no residue. The pixels of the non-residue regions are unwrapped using modified Itoh's method firstly while the residue regions have been isolated, then region-growing strategy of region-growing phase unwrapping algorithm is applied to unwrap the pixels of residue regions from the unwrapped non-residue regions. The test performed on the real wrapped phase data obtained from the interferometry of a pair SAR images shows the proposed algorithm is fast and efficient even in low-quality regions.

Image fusion is an important tool in remote sensing, since many Earth observation satellites provide both high-resolution panchromatic (Pan) and low-resolution multispectral (MS) images. To date, many image fusion techniques have been developed. However, the available algorithms can hardly produce a satisfactory fusion result for IKONOS and QuickBird images. Among the existing fusion algorithms, the IHS technique is the most widely used one, and the wavelet fusion is the most frequently discussed one in recent publications because of its advantages over other fusion techniques. But the color distortion of these two techniques is often obvious. The support value fusion technique demonstrates some advantages over the conventional methods. This study presents a new fusion approach that integrates the advantages of both the IHS and the support value techniques to reduce the color distortion of QuickBird fusion results. Different QuickBird images have been fused with this new approach. Visual and statistical analyses prove that the concept of the proposed extended fast IHS (eFIHS) and support value integration is promising, and it does significantly improve the fusion quality compared to conventional IHS (eFIHS) and wavelet fusion techniques.

Proposed is a novel approach based on independent component analysis (ICA) for speckle reduction and target extraction of SAR (synthetic aperture radar) images using adaptive space separation with weighted information entropy incorporated. First the basis and the independent components are respectively obtained by ICA technique, and Weighted Information Entropy of the image is computed, then based on the threshold computed from function T-WIE (Threshold VS. Weighted-Information-Entropy), independent components are adaptively separated and the basis are classified accordingly. Thus, the image space is separated into two subspaces -'clean' and 'noise'. Then, a proposed nonlinear operator 'ABO' is applied on each component of the 'clean' subspace for further optimization. Finally, recovery image is obtained reconstructing this subspace and target is easily extracted with binarisation. Note that here T-WIE is an interpolated function based on several representative target SAR images using proposed space separation algorithm.

Since 1986 Bayesian Network has been a hot study topic in artificial intelligence field. We have researched Bayesian Network and related algorithms in remote sensing data processing for five years. Recently we finished BayesNetEX which is an extensive edition software of Bayesian network algorithms for remote sensing image processing and knowledge inference. The copyright is from Copyright Protection Center of China. This paper briefly introduces the main modules of BayesNetEX and demonstrates its classification application with an ETM+ image, which also shows some potential applications in remote sensing image processing.

This paper studies the algorithm fusing IR/Visible images used in imaging guidance field. Firstly, the differences between IR image and Visible one are introduced curtly. Then we introduce the essential multiwavelet theory and explain the reason why we choose multiwavelet analysis as a tool fusing images. We give the framework of image fusion based on discrete multiple wavelet transform (DMWT) which core principle is still based on multiresolution analysis (MRA) and Mallat algorithm. A novel image fusion method based on multiwavelet transform and directional contrasts information is presented.

Pattern Recognition has been successfully applied to target detection. The characteristic of target pattern determines the detection ability in pattern recognition. The pattern of spectral signature at hyperspectral resolution provides more distinguished spectral feature for target detection and then improves the detection ability. However, hyperspectral image has limitation on low resolution in spatial. Therefore, this article focus on analyze on the target detection ability at the sub-pixel scale in different spatial resolution, considering two critical factors, i.e. spatial response in sensor and background interferer. Experiment data is simulated by inducing these two factors. Target-to-Clutter-Ratio (TCR²) curve and Receiver Operating Characteristic (ROC) curve with Uniform Target Detector (UTD) analyze on the simulated data. We conclude that spatial response of the sensor and the background interferer induce uncertainty into target detection ability and usually weakens it. It gives rise to a new requirement for hyperspectral target detection that should be considerate for the effect caused by spatial resolution.

In this paper, an algorithm for hyperspectral image compression is presented. It carries DCT (Discrete Cosine Transform) on spectral bands to exploit the spectral correlation and then DWT (Discrete Wavelet Transform) on every eigen image to exploit the spatial correlation. After that, 3D-SPIHT (three-dimensional Set Partitioning in Hierarchical Trees) is performed for encoding. Experiments were done on the OMIS-I (Operational Modular Imaging Spectrometer) image and the performance of this algorithm was compared with that of 2D-SPIHT. The results show that the performance of 3D-SPIHT based on DCT and DWT is much better than that of 2D-SPIHT and the quality of the reconstructed images is satisfying.

The at-surface radiances of a thermal infrared channel relates to the land materials characters, which is composed of the thermal radiance of the land itself and the atmospheric downwelling radiance reflected by the land surface. This parameter is very important for thermal infrared remote sensing, but the researches about its retrieving from the remote sensors with a wider spectral range are less. This paper discussed a method to retrieve the parameter by using the NOAA-AVHRR image itself. The method was firstly theoretically deduced based on the thermal radiance transmission theory. According to the deduction, the atmospheric transmittance and the upwelling radiance of the imaging time can be expressed by the relation of two channels at-sensor radiances, which are used to calculate the at-surface radiances. Then the simulated data were used to determine the model's coefficients and prove the feasibility of the theoretical retrieval method. The relative root mean square error for the research method was about 5.0 percent by comparing with the validation data, which showed that the method could simply and effectively compute the at-surface radiance of the two thermal infrared channels by the image itself for the AVHRR sensors. Even if lacking of the synchronous atmospheric parameters, a higher retrieval precision also exist.

Airborne laser scanning data has become an accepted data source for highly automated acquisition of digital surface models(DSM) as well as for the generation of digital terrain models(DTM). To generate a high quality DTM using LIDAR data, 3D off-terrain points have to be separated from terrain points. Even though most LIDAR system can measure "last-return" data points, these "last-return" point often measure ground clutter like shrubbery, cars, buildings, and the canopy of dense foliage. Consequently, raw LIDAR points must be post-processed to remove these undesirable returns. The degree to which this post processing is successful is critical in determining whether LIDAR is cost effective for large-scale mapping application. Various techniques have been proposed to extract the ground surface from airborne LIDAR data. The basic problem is the separation of terrain points from off-terrain points which are both recorded by the LIDAR sensor. In this paper a new method, combination of morphological filtering and TIN densification, is proposed to separate 3D off-terrain points.

Super-resolution reconstruction has been widely used in infrared images. A lot of effective super-resolution methods have been presented in recent years. In this paper, a fast and robust super-resolution algorithm based on Maximum a Posteriori (MAP) estimation is proposed to obtain a high resolution image from a set of infrared images, which are obtained by an uncooled infrared detector. A comparison and an analysis are made of the super-resolution reconstruction results by this method, with the variance of regularizations and the number of low resolution infrared images, by direct observation and the value of Power Signal-to-Noise Ratio (PSNR). Simulation results with several real sets of infrared images show the effectiveness and superiority of this method for enhancing resolution of infrared images.

In this paper, we present a new iterative blind multispectral image restoration algorithm based on double regularization (DR). The motivation for DR when applied to multispectral restoration lies in its effectiveness towards edge preservation in joint blur identification and image restoration. With consideration for both the intra- and inter-channel blurring function in the multiple-input multiple-output (MIMO) systems, an alternating minimization (AM) procedure with conjugate gradient optimization (CGO) scheme is formulated to implement restoration iteratively. The derivation of DR optimization shows that optimal restoration result can be achieved even when the MIMO systems suffer from inter-channel interference. Experimental results show that it is effective in performing blind mutichannel restoration when applied to color images.

This paper developed a new model of region extraction based on salient region detection and scale-space primal sketch. In the proposed model, we extract the region of interest (ROI) in two steps. Firstly, we estimate the extent of object by means of region detection, which considers the feature that contributes most to the saliency map. Secondly, we use the scale-space primal sketch to acquire an explicit representation of the significant image structure which gives a qualitative description of the scales and regions of interest. Finally, we combine the results from the two steps. Applications to extract ROI showed that this new model could lead to better results which can be used for guiding later stage processing.

In this paper, a new image fusion frame based on nonsubsampled contourlet transformation is provided. In this frame, a new method based on the local correlation coefficients is explored, the experiment results of this method can obtain relatively high spatial resolution and preserve high spectral resolution.

Image fusion based on wavelet transform is the most commonly used image fusion method, which decomposes the source images, fuses their coefficients according to some fusion rules and then reconstructs the fused image. Its two main traditional rules are selecting maximum absolute value and the combination of selecting and weighted averaging. Both of the two rules did some artificial supposes to eliminate the uncertainty of the extent of each source image's contributions, so they both ignored some useful information and were sensitive to noise. Fuzzy reasoning is the best way to resolve uncertain problems. As a result, this paper proposed a new image fusion algorithm based on wavelet transform and fuzzy reasoning. It first decomposed source images through wavelet transform, computed the extent of each source image's contribution through fuzzy reasoning using the area feature of source images' wavelet coefficients, and then fused the coefficients through weighted averaging with the extents of each source images' contributions as the weight coefficients. Finally it did inverse wavelet transform to produce the fused image. Using the mutual information and PSNR as criterions, experiment results demonstrated that the new algorithm was more effective and robust than the traditional fusion algorithms based on wavelet transform.

We present an effective image recovering method .Edge information are very important, we can detect the edge of the image via wavelet transform before recovering the image. In this paper, we give a simple and effective method to detect the edge information. The first, we can decompose the image using a wavelet transform, the high frequency information is corresponding to the edge and noise, the edge and noise have different properties, as follows: the orientation of edge is very strong, but the noise is non-orientation. On base of that, we can decompose the edge and noise in horizontal orientation, vertical orientation, and diagonal orientation. Because the edge is directional, its values in each orientation have much difference; in opposition, noise is non-orientation, its values are almost same. We can set the threshold based on the variances of difference in each orientation. By this time, we gain the edge without noise. For the rest, we can restore using regularization. Now the low frequency is corresponding to the flat. In this paper, we detect the edge by the wavelet, so can choose a bigger parameter to recover. At last, we add the edge and the part of regularized image restoration. This method has advantage in holding the edge and is simple to choose the parameter of regularization. Experimental results show the good performance, this method can keep image's of edge from degradation and increase PSNR up to 1~2dB.

An improved local RX anomaly detection algorithm is proposed. It firstly projects the images onto the background orthogonal subspace to make local data closer to multivariate normal distribution. Then for every tested pixel in the center of the sliding local window, the bands used in RX detector are chosen adaptively. To avoid the influence of anomaly information on the background characteristic statistic, the anomalous pixels in the local background are removed and the covariance matrix is calculated using real background pixels. Finally the RX detector is used to calculate the anomalous degree of every tested pixel. Experimental results indicate it is robust and has good anomaly detection performances under complex unknown background.

In this paper, we propose a new de-noising method by wavelet transform based on lifting scheme for the reducing of the multiplicative speckles in synthetic aperture radar (SAR) images. An ERS-2 SAR image of Hangzhou was used as a test image to compare the performance of the method with that of conventional methods. The results show that the proposed method has advantages in radiation characteristics and textual details of the image over the conventional methods.

Color fusion algorithm for visible and infrared(IR) images based on color transfer in YUV color space under trees, lawn, or land background is presented. Considering the red color will alert observers to possible interested target or danger, this paper aims at working on an algorithm that emphasizes hot targets in IR image with intense red, and the background details in visible image present natural color similar to a color day-time image. V component of YUV space represents the difference between red and Y. Properly increasing the V value will obtain intense red color. Therefore, a nonlinear transfer method based on local mean value of the IR image is proposed. A window of size 5x5 is used to locate hot target in IR image. When the local gray mean value in this window is larger than the global mean value, we determine that this pixel is in a hot area. Then its V value is increased by the ratio of the local gray mean value to the global mean value. Tests show that this method pops out the hot targets with intense red color while the background rendered natural color appearance.

Aiming at the problem of negative index in the spectral color space built by means of traditional principal component analysis (PCA), a method of color component prediction based on rotated principal component analysis (RPCA) is proposed, which performs the rotating transformation from initial eigenvectors to a set of all-positive vectors as the physical basis color components while retaining the cumulative ratio of the variance contributions of significant principal components to the original multispectral space to the maximum extent. The rotated column vectors should be also polarized between 0 and 1. The spectral database of Munsell Matte Collection I is used for experiment. The experimental results show that the novel method of prediction not only uncovers the real color components of the target image better but reconstructs the normalized spectra data set with a high colorimetric and spectral accuracy. Thereinto, the colorimetric errors of the four estimated components reconstruction for more than 96 percent of the samples in Munsell Matte Collection I are less than 3 units of color difference acceptable.

Ocean features in synthetic aperture radar (SAR) images are usually complicated and make SAR images hard to understand. Because of lower signal-to-noise rate (SNR) in SAR imagery, it is much more difficult to separate different ocean features than to separate different land features. This paper presents a completely novel method to separate ocean features from multi-band polarimetric SAR imagery based on polarimetric signatures of ocean features. AIRSAR data from Jet Propulsion Laboratory (JPL), National Aeronautics and Space Administration (NASA) are used in the case studies and good results are achieved.

In this paper, we propose a new method to generate complex IR scene image directly from the corresponding visual scene image based on material thermal database. For the input visual scene image, we realize an interactive tool based on the combined method of global magic wand and intelligent scissors to segment the object areas in the scene. And the thermal attributes are assigned to each object area from the thermal database of materials. By adopting the scene infrared signature model based on infrared Physics and Heat Transfer, the surface temperature distribution of the scene are calculated and the corresponding grayscale of each area in IR image is determined by our transformation rule. We also propose a pixel-based RGB spacial similarity model to determine the mixture grayscales of residual area in the scene image. To realistically simulate the IR scene, we develop an IR imager blur model considering the effect of different resolving power of visual and thermal imagers, IR atmospheric noise and the modulation transfer function of thermal imager. Finally, IR scene images at different intervals under different weather conditions are generated. Compared with real IR scene images, our simulated results are quite satisfactory and effective.

Terrain aided navigation (TAN) is an efficient way to periodically correct the error accumulation of INS. The imaging laser radar is an ideal imaging sensor in TAN for the low-flying aircraft and unmanned air vehicles for the high precision multi-dimensional data acquisition capability and concealable attribute. In this paper, a new framework for applying the laser radar to terrain aided navigation is put forward. Then a new adaptive fused Kalman Filter is proposed to improve the accuracy and robustness. At last, the key factors affected the algorithm are analyzed and the comparative experimentations are presented. The simulating experiments show that the proposed algorithm improves the location accuracy, and has good initial error tolerance and fine robustness. It shows that this approach is a valid solution for the application.

When designing the system of the infrared (IR) small target detection, both which detection algorithm to be chosen and how to set some parameters of the detection algorithm need to evaluate the complexity of the infrared backgrounds. In this paper, an effective method to evaluate the complexity of the sea-sky infrared backgrounds based on texture analysis is proposed. According to the characteristics of Infrared backgrounds, three statistical descriptors based on gray level co-occurrence matrix are calculated, and one statistical descriptors based on edge information is calculated. All statistical descriptors are fused to evaluate the complexity of IR backgrounds finally. Real and synthetic IR images under sea-sky background are applied to validate the proposed approach. Compared to existing methods, Experimental results demonstrate the robustness of the proposed method with high performance.

With the extensive application of mobile robots in many different fields, map building in unknown environments has been one of the principal issues in the field of intelligent mobile robot. However, Information acquired in map building presents characteristics of uncertainty, imprecision and even high conflict, especially in the course of building grid map using sonar sensors. In this paper, we extended DSmT with Fuzzy theory by considering the different fuzzy T-norm operators (such as Algebraic Product operator, Bounded Product operator, Einstein Product operator and Default minimum operator), in order to develop a more general and flexible combinational rule for more extensive application. At the same time, we apply fuzzy-extended DSmT to mobile robot map building with the help of new self-localization method based on neighboring field appearance matching( -NFAM), to make the new tool more robust in very complex environment. An experiment is conducted to reconstruct the map with the new tool in indoor environment, in order to compare their performances in map building with four T-norm operators, when Pioneer II mobile robot runs along the same trace. Finally, a conclusion is reached that this study develops a new idea to extend DSmT, also provides a new approach for autonomous navigation of mobile robot, and provides a human-computer interactive interface to manage and manipulate the robot remotely.

Contourlet Transform is efficient for image representation due to its capability of capturing smooth contours in natural images and hence its ability for natural images denoising is promising. Because the transform is not orthogonal, the shrinkage factors for images denoising with contourlet transform method is always estimated by Mento-Carlo method which is much time consuming and the recovered images are always somewhat blurred especially in high intensity noise conditions. To overcome the two deficiencies, we proposed a modified approach which we call subband effect factors method. Finding the factors is less time cost and it's much more efficient for images denoising than Mento-Carlo method by contourlet transform. Using the factors, we present a hard threshold denoising method by modifying the 3σ rule utilizing subband threshold effect factor of each subband in second version contourlet domain. The threshold effect factor of every subband is acquired according to the characteristics of every subband response driven by a normalized white Gaussian noise image. Experimental results show that, for natural and SAR images corrupted by Gaussian white noise, the denoising results including PSNR (peak signal-noise ratio) and visual quality are superior to those of Mento-Carlo method, especially when the noise is high intensity.

In the past few years, wavelet-domain hidden Markov models have proven to be useful tools for statistical signal and image processing. The hidden Markov tree (HMT) model captures the key features of the joint probability density of the wavelet coefficients of real-world data. One potential drawback to the HMT framework is the deficiency for taking account of intrascale correlations that exist among neighboring wavelet coefficients. In this paper, we propose to develop a joint hidden Markov model by fusing the wavelet Bayesian denoising technique with an image regularization procedure based on HMT and Markov random field (MRF). The Expectation Maximization algorithm is used to estimate hyperparameters and specify the mixture model. The noise-free wavelet coefficients are finally estimated by a shrinkage function based on local weighted averaging of the Bayesian estimator. It is shown that the joint method outperforms lee filter and standard HMT techniques in terms of the integrative measure of the equivalent number of looks (ENL) and Pratt's figure of merit(FOM), especially when dealing with speckle noise in large variance.

Synthetic aperture radar (SAR) images are inherently affected by a signal dependent noise known as speckle, which is due to the radar wave coherence. In this paper, we propose a novel adaptive despeckling filter and derive a maximum a posteriori (MAP) estimator for the radar cross section (RCS). We first employ a logarithmic transformation to change the multiplicative speckle noise into additive noise. We model the RCS using the recently introduced Generalized Gaussian density function^[1], Which was proved to be the best described for the SAR Amplitude. We estimate model parameters from noisy observations by means of second-kind statistics theory, which relies on the Mellin transform. Finally, we compare the proposed algorithm with several classical speckle filters applied on actual SAR images. Experimental results show that the MAP filter based on the Generalized Gaussian prior for the RCS is among the best for speckle removal.

Urban is a kind of very complex living space, and includes many and many different objects. For urban imagery, it is hardly to get a satisfied classification result utilizing any kind of single classification method. Aiming at this problem, this paper adopts a kind of classification method based on stepping masking principle using the parallel-pipeline classification and the improved FCM method. With this classification principle, it not only enhances the computation efficiency of classification, but also realizes the accurate and reasonable classification to the different kinds of urban objects. Finally this paper evaluates the precision of classification results using the confusion matrix and the Kappa coefficient separately. Analyzing from the classification effect and precision, this algorithm can satisfy the requirement of classification in different level or the thematic mapping.

A new method for estimating noise in hyperspectral images is described in this paper. It is based on the strong between-band correlation of hyperspectral images and the concept of local standard deviations of small imaging blocks. The new method can be used to automatically estimate noise for both radiance and reflectance images. Unlike other methods discussed in this paper, the new method is more reliable for estimating noise in hyperspectral images with diverse land cover types. We successfully applied the new method in estimating noise for Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data.

This paper describes an algorithm framework for registration of airborne based laser scanning data (LIDAR) and optical images by using multiple types of geometric features. The technique utilizes 2D/3D correspondences between points and lines, and it could easily be extended to general features. In generalized point photogrammetry, all lines and curves are consists of points, which could be describe in collinear equation, so it could represent all kinds of homogeneous features in an uniform framework. For many overlapping images in a block, the images are registered to the laser data by the hybrid block adjustment based on an integrated optimization procedure. In addition to the theoretical method , the paper presents a experimental analysis the sensitivity and robustness of this approach

A novel 3D terrain matching algorithm is presented in this paper. A terrain feature vector map (FVM), composed of local mean and local gradient, is employed to represent the terrain elevation map (TEM). Compared with traditional matching algorithm using the magnitude of gradient to match, the new algorithm uses each component of the gradient vector to match individually, and it is able to generate two interim matching positions. Different from traditional matching algorithms which usually estimate an optimum matching position under some criterions at the end, the new algorithm fused the two interim matching positions to generate a final matching position or refuse to position in order to increase the matching confidence, which is very important because it is hardly acceptable to employ a mismatched position to correct the error of Inertial Navigation System (INS). Due to the stability of terrain and the high-precision of lidar ranging, the mean of a sensed terrain elevation map (STEM) sized terrain is quite stable. So it is bestowed to accelerate the matching process and to reduce mismatches at different terrain heights. Compared with other mismatch-eliminated methods based on neural network (NN) or support vector machine (SVM), the new method do not need training samples and is more stable and robust. Experimental results show that the proposed algorithm is effective and robust.

In this paper, a nonlinear model of nonuniformity correction (NUC) algorithm based on Kalman Filter is adopted and improved by using piecewise method of detector response, which is on the premise that response factors of each detector in infrared focal plane arrays (IRFPA) are regarded as random state variables modeled by a discrete-time Gauss-Markov process. The effect of nonlinear response of the IRFPA detectors to the NUC accuracy can be solved by our algorithm effectively. The algorithm can also improve the NUC result, especially, in the condition of wide dynamic range application. The experimental results show that our algorithm gives good correction precision and performance.

This paper proposes a global threshold selection method to do infrared image segmentation, which uses both gray-level distribution and spatial information, namely, two-dimensional OTSU method (2D OTSU). It often gets better anti-noise performance. What's more, taking consideration of the complexity of its computation, we introduce a new heuristic optimization algorithm, called the particle swarm optimization (PSO) algorithm to search the result. So an algorithm for PSO-based 2D Otsu segmentation is proposed. The experiments of segmentation the infrared images are illustrated to show that the proposed method can get ideal segmentation result with less computation cost.

Estimates of vegetation water content are of great interest for assessing vegetation water status in agriculture and forestry, and have been used for drought assessment. This study focuses on the retrieval of foliar water content with hyperspectral data at canopy level. The hyperspectral image used in this study was acquired by the airborne operative modular imaging spectrometer (OMIS) at Demonstration Site for Precision Agriculture in Xiaotangshan area, Beijing, on April 26th, 2001. 40 image spectra were extracted to correspond to the quasi-synchronous meansurements of foliar water content (FWC). The image spectra of winter wheat were utilized to validate the sensitivity of the existing and novel water indices and parameters of three water absorption features in NIR and SWIR regions. Correlation analysis showed that, NDWI(860,1241) and NDWI(860,1200) both had significant linear relationships with FWC (R² were 0.4124 and 0.4042 respectively). Red edge position (REP) could reflect indirectly the variations of wheat FWC to some extent. Significant linear relationships were also found between WI(820,1600) and FWC, and between WI(900,1200) and FWC, while no relationship was shown between the traditional WI(900,970) and FWC. The derived depth of water absorption centered around 2078nm, namely AD2078, had the highest linear correlation with FWC (R2 is 0.4551) , much higher than those parameters derived from the two water absorption around 1175 and 1409. In the end, AD2078 was applied to OMIS image to map the foliar water content. The value range of the inverted foliar water content ranged from 69.39 to 78.35%, which was quite close to that of the field measurements (70.72-78.12%). The distribution of the FWC map was quite consistent with growth status of winter wheat.

Beijing-1 satellite is a DMC+4 microsatellite which will be used in land use, city planning field. It provides both multispectral (MS) and panchromatic (PAN) data with spatial resolutions of 32m and 4m, respectively. Fusion is used to produce high-resolution multispectral images from a PAN image and low-resolution MS images. In this paper, beijing-1 PAN image and MS images of Chengdu are fused by both COS method and wavelet transform method. Several experiments are taken to test the properties of fusion result. Experiment results show that wavelet based fusion method provided satisfied results with better quality than the other methods in both spatial and spectral domains.

Coherent Target Analysis (CTA) method derived for the (Interferometric SAR) InSAR data processing has developed as one of the critical techniques in the field of surveying subtle land deformation. This paper discusses the principles and characteristics of CTA procedure, which can remove the errors caused by temporal and spatial decorrelation and atmospheric effects. The paper presents the validation of characteristics of CTA applied in city subsidence in Shanghai test site, where has been affected by subsidence for more than eighty years. The 32 scenes from the ERS-1/2 satellites covering the test area between 1992 and 2000 were chosen and processed. Based on the theory of CTA, we can get the map of land subsidence of Shanghai. Quality assessment between the velocity fields derived of CTA and that of leveling data observed on benchmarks within the same periods is then discussed. The comparison with ground leveling data shows the derived result from CTA approach is consistent with leveling data. The further investigation of CTA approach in Shanghai will be continued to improve the accuracy and robustness.

An iterative blind deconvolution algorithm for degrade image is presented in this paper. The algorithm includes two steps, namely, the estimation of the point spread function of degrade image and the restoration using estimated point spread function. Two different Hopfield neural networks are built for realizing the two steps. An iterative procedure is used to control the restoration process. The simulation results indicate that the method is effective for blind deconvolution with high convergence speed.

A non-parametric random field model by Ashikhmin algorithm is introduced to synthesize the infrared textures in detail. The non-parametric model provides the spatial structure of the infrared texture, which makes it more similar with the infrared textures taken by infrared camera. Then a conversion of texture to temperature field is given to yield a temperature distribution by given a group of temperature parameters. The Planck's radiation law is applied to obtain the emittance distribution sample, called physical infrared texture which is always used in the hardware-in-the-loop systems, from the temperature field. And a method to generate the appearance infrared texture is developed by applied the effect of infrared imaging system. Finally, some examples of infrared textures synthesized by Ashikhmin algorithm are presented in the paper. The simulation results show that the non-parametric model is more available, more efficient, more suitable and more easily controlled to synthesize the infrared textures in the infrared scenes simulations.

With the availability of multi-sensor data in the field of remote sensing, sensor fusion has emerged as a promising research area. This study presents a simple spectral preservation fusion approach based on band ratio and weighted combination. It injects spatial features into multi-spectral images to improve the spatial information, and adjusts the ratio between the high spatial resolution image and the multi-spectral image with a weight factor to reduce the color distortion. This method is applied to merge SPOT and LANDSAT (TM) images. Visual and statistical analysis prove that the technique presented here is clearly better than the conventional image fusion techniques for preserving the spectral properties with the spatial detail improved synchronously.

This paper presents a novel fusion approach using PCA merger based on multiscale decomposition (MSD), combined with region segmentation and support vector machine (SVM), the result is a high spatial resolution multispectral image from a high resolution panchromatic (Pan) image and low resolution multispectral (Ms) images. Principal components analysis (PCA) fusion technique is one of typical fusion methods, and PCA merger based on MSD had been proposed which can obtain better performance. As we know that, in pixel fusion level, the original images are fused as internal region regardless of the contents of images, but in this paper, we perform region segmentation after MSD, because the homogeneous regions have similar features such as color, texture and intensity. Traditionally, various fusion rules can be applied after MSD according to different conditions, however, the crucial problem is which fusion rule should be adopted under given condition, hence we use the SVM to combine the most fusion rules so that can avoid some drawbacks using single fusion rule. To validate our approach, we compare it with several typical fusion approaches, and the best result is obtained using our approach.

Laser radar can simultaneously produce the range image and the intensity image, and it can directly collect rich information of target. Compared with the other sensors, such as infrared or radar, laser radar can enhance the recognition rate and the precision of target aimed point. When laser radar vertically detects the objects on the plane ground, the correlation filters with in-plane rotation invariance are usually used to solving the problem of the target recognition. Traditional correlation filters are still improved on the aspect of recognition rate. In the paper, through deducing the relationship between support vector machine (SVM) and correlation principle in the signal processing, a new correlation filter, named linear SVM correlation filter (LSCF) that has the properties of SVM, is proposed. The real images of laser radar are used as the training and testing samples. The experiments state that the filter has good recognition attributes, such as stable correlation output and high recognition rate. LSCF is suitable to be the recognition algorithm of the imaging laser radar.

This paper presents a fusion method for multi-focus images to produce a new clear image for the scene. First we analyze the defocused image formation process and obtain a set of ill-posed equations. By imposing some constraints, they become well-posed and the fusion task is converted to a simple optimization problem. Then the optimization problem, which minimizes the gradient difference and intensity difference with respect to the objective gradient field and intensity constraints, is equivalent to the optimal fusion method for multispectral image fusion. So we construct the objective gradient field and the intensity constraints by clarity analysis and model change respectively and obtain the minimization result by the iterative optimization steps of optimal fusion. At last, the experiments convincingly demonstrated that the proposed method has better tolerance to misalignment and noises than wavelet fusion.

A direct change detection method that utilizes the dynamic Bayesian network (DBNs) is proposed to detect buried objects. The DBNs uses the time series dynamic data to produce credible probabilistic reasoning, and is developed to utilize the IR images obtained by different band and temporal. The proposed method offers a way to change detection analysis from the static viewpoint to the dynamic viewpoint, which can input and deal with more than two multi-temporal images simultaneously which are featured by multi-band. The origin of thermal contrast in infrared imaging between the buried objects and background is illuminated on the theory of infrared radiation. The differences of temperature can be captured by multi-temporal and multi-band infrared images. The IR images of the regions of interest (ROI) acquired at three different times as inputs to detect buried objects using multi-temporal direct change detection based on physical principle of infrared imaging. The experimental results indicate that the change detection method based on DBNs is an effective to buried objects detection.

This paper deals with subpixel accuracy Synthetic Aperture Radar (SAR) image registration which also satisfies real-time demand of SAR image processing. Because of the presence of speckle noise in SAR image, improved region detection method are carried on in reference image and sensed image respectively firstly. Then each region is represented by a set of invariant moments and chain coding of the region boundary. Correspondence between the regions in the reference image and sensed image is established by the improved regions matching criterion which proposed by us. The centers of gravity and the corners on the region boundary are the potential control points. Correspondence between the control points is established in the feature space, using the principle of minimum distance classifier. After finding enough and right matched control points, interpolation and estimation of transformation parameters using least squares method are executed finally.

Since Chavez proposed the highpass filtering procedure to fuse multispectral and panchromatic images, several fusion methods have been developed based on the same principle: to extract from the panchromatic image spatial detail information to later inject it into the multispectral one. In this paper, we present new fusion alternatives based on the same concept, using the multiresolution contourlet decomposition to execute the detail extraction phase and the generalized intensity-hue-saturation (GIHS) and principal component analysis (PCA) procedures to inject the spatial detail of the panchromatic image into the multispectral one. Experimental results show the new fusion method have better performance than GIHS, PCA, wavelet and the method of improved GIHS and PCA mergers based on wavelet decomposition.

In passive millimeter wave (PMMW) imaging, antenna size limitations lead to the problem of poor resolution of acquired image. Thus efficient post-processing is necessary to achieve resolution improvement. In this paper, we present an Adaptive Porjected Landweber super-resolution algorithm that attempts to leverage the strong points of both Landweber iteration and projection-based adjustments. In the algorithm, we implement the Landweber iterations as the main image restoration scheme and include a projection-based adjustment for enforcing constraints after each Landweber iteration is completed. Furthermore, the algorithm updates the parameter adaptively at each iteration. From experiments, we find that the Adaptive Projected Landweber superresolution algorithm obtains better results and has lower mean square error (MSE) and produces sharper images. These constraints and adaptive characters speed up the convergence of the Landweber algorithm.

Recognition of targets in high-resolution synthetic aperture radar(SAR) imagery is a challenging problem in practice. Features of military targets in SAR imagery play important roles in SAR ATR. Robust feature extraction of these targets is difficult due to extended operating conditions such as obscuration, articulation, varied configurations and a host of camouflage. In this paper, a new method based on wiener filtering reconstruction in multi-wavelet domain to enhance target region feature is presented. It uses a wavelet thresholding estimate as a mean to design a wavelet domain Wiener filter employing two different bases of wavelet, one for the thresholding step and another for the filter application. The experimental results demonstrate that the method can improve the target regional feature, can augment statistical separability between targets and clutter, and benefits to speckle suppression using publicly released SAR data from DARPA's MSTAR program.