The partitioning of an image may be defined as the division of its image plane into image objects. This paper presents a new methodology for partitioning multispectral images. It combines the analysis of the spectral properties of the pixels with the analysis of their spatial properties. Spectral properties are studied in the multivariate histogram of the image, while spatial properties are analyzed in the multispectral gradient of the image. The histogram of the image is first segmented by a nonparametric algorithm. The segmented histogram allows the classification of all image pixels. Each resulting class is then separately filtered in order to remove all classified pixels having a high probability of being misclassified when considering spatial criteria. The filtered classes are used as seeds for boundary detection on the gradient of the original image. The class of the resulting regions is given by the class of the seeds that created those regions.

A parameterized t-opening is a filter defined as a union of openings by a collection of compact, convex structuring elements, each scalar multiplied by the parameter. For a reconstructive t-opening, the filter is modified by fully passing any connected component not completely eliminated. Applied to the signal-union-noise model, in which the reconstructive filter is designed to sieve out clutter while passing the signal, the optimization problem is to find a parameter value that minimizes the MAE between the filtered and ideal image processes. The present study introduces an adaptation procedure for the design of reconstructive t-openings. The adaptive filter fits into the framework of Markov processes, the adaptive parameter being the state of the process. There exists a stationary distribution governing the parameter in the steady state and convergence is characterized via the steady-state distribution. Key filter properties such as parameter mean, parameter variance, and expected error in the steady state are characterized via the stationary distribution. The Chapman-Kolmogorov equations are developed for various scanning modes and transient behavior is examined.

The theory of an image decomposition that we refer to as a sieve is developed for images defined in any finite number of dimensions. The decomposition has many desirable properties including the preservation of scale-space causality and the localization of sharp-edged objects in the transformation domain. The decomposition has the additional properties of manipulability, which means that it is easy to construct pattern recognition systems, and scale-calibration, which means that it may be used for accurate measurement.

Soft morphological filters form a large class of nonlinear filters with many desirable properties. However, few design methods exist for these filters. This paper demonstrates how optimization schemes, simulated annealing and genetic algorithms, can be employed in the search for soft morphological filters having optimal performance in a given signal processing task. Furthermore, the properties of the achieved optimal soft morphological filters in different situations are analyzed.

A filter structure formed as a linear combination of stack filters (L-stack-filters), is studied. This type of filters include many useful filter classes, e.g., linear FIR filters and nonlinear threshold Boolean filters, and L-filters. An efficient algorithm for finding optimal filter coefficients under the mean squared error (MSE) criterion is derived. A subclass of the above filters, called FFT-ordered L-filters (FFT-LF), is studied in detail. In this case the bank of filters is formed according to a generalized structure of the FFT flowchart. It is shown that FFT-LFs effectively remove mixed Gaussian and impulsive noise. While possessing good characteristics of performance, FFT-LFs are simple in implementation. In the sense of implementation the most complicated FFT-LFs are the well-known L-filters. We suggest an efficient parallel architecture implementing FFT-LFs as well as a family of discrete orthogonal transforms including discrete Fourier and Walsh transforms. Both linear and nonlinear L-filter-type filters are implemented effectively on the architecture. Comparison with known architectures implementing both linear and nonlinear filters reveals advantages of the proposed architecture. An efficient implementation of L-stack-filters is also proposed.

Mathematical morphology on sets can be understood as a formal language, whose vocabulary comprises erosions, dilations, complementation, intersection and union. This language is complete, that is, it is enough to perform any set operation. Since the sixties special machines, called morphological machines (MMachs), have been built to implement this language. In the literature, we find hundreds of MMach programs that are used to solve image analysis problems. However, the design of these programs is not an elementary task. Thus, recently much research effort has been addressed to automating the programming of MMachs. A very promising approach to this problem is the description of the target operator by input-output pairs of images and the translation of these data into efficient MMach programs. This approach can be decomposed into two equally important steps: (1) learning of the target operator from pairs of images; (2) search for economical representations for the operators learned. The theory presented in this paper is useful in the second step of this procedure. We present some set operations on collections of closed intervals and give efficient algorithms to perform them. These operations are used to parallelize MMach programs and to prove the equivalence between distinct MMach programs.

In the unsharp masking approach for image enhancement, a fraction of the highpass filtered version of the image is added to the original image to form the enhanced version. The method is simple, but it suffers from two serious drawbacks. First, it enhances the contrast in the darker areas perceptually much more strongly than that in the lighter areas. Second, it enhances the noise and/or digitization effects, particularly in the darker regions, resulting in visually less pleasing enhanced images. In general, noise can be suppressed with lowpass filters, which are associated with the blurring of the edges. On the other hand, contrast can be enhanced with highpass filters, which are associated with noise amplification. A reasonable solution, therefore, is to use suitable nonlinear filters which combine the features of both highpass and lowpass filters. This paper outlines several new methods of unsharp masking based on the use of such nonlinear filters. Computer simulations have verified the superior results obtained using these filters. In addition, a new measure of contrast enhancement is introduced which quantitatively supports the improvement obtained using the proposed methods.

This paper addresses a nonlinear multiresolution decomposition scheme based on order statistic (OS) filters with applications in image restoration. In this scheme the input signal is decomposed into multiresolution bands using OS filters of different parameter values. We present the deterministic and statistical properties of the decomposed signals. As an application of the proposed scheme, we consider the restoration of images in the presence of Gaussian and non-Gaussian noise. In the image restoration, nonlinear filtering is performed on decomposed signal components using denoising techniques. Simulations show that this method has better edge and sharp feature preservation and better noise reduction properties than other OS filter based techniques.

A technique that can improve the optical character recognition (OCR) accuracy of text images is presented. By using the output from an OCR system and a distorted text image, this technique trains an adaptive restoration filter and then applies the filter to the distorted text that the OCR system could not recognize. The restored text image is then reprocessed by the OCR system, and the restored characters are recognized with a higher accuracy than the distorted text. A series of experiments were performed to determine a feasible adaptive restoration filter architecture, to establish this technique's ability to restore uniformly distorted text, and to demonstrate this technique's ability to improve the OCR accuracy of real world text documents. The results of these experiments show that this technique can improve both pixel and OCR accuracy of distorted text images.

This paper presents a generalized image contrast enhancement technique, which equalizes the perceived brightness distribution based on the Heinemann contrast discrimination model. It is based on the mathematically proven existence of a unique solution to a nonlinear equation, and is formulated with easily tunable parameters. The model uses a two-step log-log representation of luminance contrast between targets and surround in a luminous background setting. The algorithm consists of two nonlinear gray scale mapping functions that have seven parameters, two of which are adjustable Heinemann constants. Another parameter is the background gray level. The remaining four parameters are nonlinear functions of the gray-level distribution of the given image, and can be uniquely determined once the previous three are set. Tests have been carried out to demonstrate the effectiveness of the algorithm for increasing the overall contrast of radiology images. The traditional histogram equalization can be reinterpreted as an image enhancement technique based on the knowledge of human contrast perception. In fact, it is a special case of the proposed algorithm.

This study investigates a new approach to removing noise that preserves the fine details which are based on combination of mathematical morphology and fuzzy reasoning. We extract some features from an image using mathematical morphology, then input these features, and the image itself, into fuzzy-rule-based systems to produce a more graceful image.

Car identification is a goal in traffic control, transport planning, travel time measurement, managing parking lot traffic and so on. Most car identification algorithms contain a standalone plate segmentation process followed by a plate contents reading. A pyramidal algorithm for license plate segmentation, looking for textured regions, has been developed on a PC based system running Unix. It can be used directly in applications not requiring real time. When input images are relatively small, real-time performance is in fact accomplished by the algorithm. When using large images, porting the algorithm to special digital signal processors can easily lead to preserving real-time performance. Experimental results, for stationary and moving cars in outdoor scenes, showed high accuracy and high scores in detecting the plate. The algorithm also deals with cases where many character strings are present in the image, and not only the one corresponding to the plate. This is done by the means of a constrained texture regions classification.

Novel adaptive robust filtering algorithms applicable to radar image processing are proposed. They take into consideration the peculiarities of radar images and possess a good combination of properties: effective speckle suppression, impulsive noise removal, edge and detail preservation and low computational complexity. The advantages of these digital algorithms are demonstrated by simulated data and images obtained by airborne side-look non SAR radar.

The title of this book is broad enough for a series of books. So I began reading this book with the preconception that it would not be sufficiently deep to be of any more value than the other image processing books I already own. I was pleasantly surprised. One measure of a book's value is how often a person reaches for it. I have reached for this book many times and have usually found it a source of useful information or leads to more in-depth sources.