Filtros espaciais: implementação, estudo comparativo e aplicação em sensoriamento remoto.

Abstract

Spatial filters have been increasingly used in enhancement and noise removal tasks in digital images. This work describes the implementation of a modular software package for spatial image filtering and a bibliographical update of current techniques in digital image processing. A comparative study between the techniques of softening established in the literature and techniques recently developed and published was carried out. An application of methods based on selective and adaptive neighborhood criteria in the pre-processing of multispectral images is presented. The current algorithms in the literature were studied and implemented in C language. A digital image processing tool, called FILTRIX, was created and dedicated to the SITIM family of microcomputer-based image processing systems. research institutions. The package includes low-pass filters, high pass and statistical and graphical tools to evaluate the performance of the algorithms in the various applications. With the rapid development of digital image processing techniques and the large expansion of application areas, adaptive and efficient filtering methods have been proposed. A comparative study was carried out between various smoothing techniques, aiming to measure their performance in noise removal operations, impulse edge preservation, impulsive noise removal, preservation of faint details and immunity to shape distortion. Techniques consecrated in the literature and techniques recently developed and published have been considered. For performance evaluation, measurements of standard deviation in homogeneous area, mean square error, scanning profiles and visual interpretation were used. The results showed the great efficiency of the Sigma filter and the algorithms based on the criterion of selective neighborhood. Further experiments using scanned and synthetic images were performed to illustrate the homogenization effects produced by low pass filters and the sensitivity of edge detection methods in gradient imaging. In this work, a method for generating spatial attributes in multispectral images is also presented through the use of nonlinear spatial filtering techniques. Preliminary results are shown. The adjusted order filter and the neighborhood smoothing selected by sum of the absolute differences were applied to an image obtained by the LANDSAT TM / 5 satellite of WRS coordinates 223 .76. The algorithm of selection of attributes based on the criteria of maximum and minimum JM distances was used to obtain a composition with 3 channels, more discriminant for the classes of interest. The image was classified by the maximum likelihood method. As a result, an increase in accuracy was obtained in the classification of compositions containing pre-processed bands relative to the classification with the original composition.