[Show abstract][Hide abstract]ABSTRACT: Image filtering is a stage commonly implied in remote sensing data processing in order to remove noise. In this paper we pay attention to filtering the images formed by millimeter (Ka) band side look aperture radars (SLARs) that can serve as imaging subsystems of multipurpose airborne remote sensing complexes. For SLAR images the dominant factor degrading their quality is multiplicative noise that is characterized by probability density function close to Gaussian and relative variance. The goal of this paper is to study how the errors in multiplicative noise variance evaluation influence the performance of different filters with application to Ka-band SLAR image processing. We have considered two test images, one containing a lot of texture and details, and the other image has many homogeneous regions. If the test image contains a lot of details and texture, the optimal value is commonly less than for test image that contains a large percentage of homogeneous regions. The error of variance evaluation more strongly influences the performance of the modified sigma filter than the performance of the local statistic Lee and DCT-based filters. The performance of filters greatly depends upon considered test image, noise statistical characteristics, setting the filter parameters. The numerical simulation data presented serve as good background and motivation for the design of locally adaptive filters that perform hard or soft switching of several different filter outputs in order to make use of the advantages of these filters and to avoid their drawbacks.
[Show abstract][Hide abstract]ABSTRACT: MM-band images formed by side-look aperture radars (SLARs) contain a lot of useful information concerning properties and characteristics of the sensed terrain. But they are commonly degraded by multiplicative noise. Sometimes, in case of image transferring from the imaging system carrier via communication line, impulse bursts can also take place. Thus, in order to improve useful information retrieval, the image preprocessing is required, and quite often it has to be performed in real time or, at least, rather quickly. For the considered application and noise properties, one approach to image processing can be the use of multistage procedures. At the first stage, the multiplicative noise variance can be evaluated (this stage is optional and can be skipped if statistical characteristics of noise are a priori known). Then we propose to apply the specially designed method for detection and removal of impulse bursts. Finally, the three-state texture/detail preserving hard-switching locally adaptive filtering (LAF) of pre-processed image is to be performed that serves suppression of remained multiplicative noise. The main goal of this article is to concentrate on fast hardware-software realization of the proposed multistage filtering procedure. Since we ought to provide the ability to operate in real time it is reasonable to use specialized hardware units that implement basic algorithm blocks. Thus, the pipeline architecture is the best choice for providing optimal computation load of the hardware units used for ensuring the highest overall processing speed for the proposed multistage filter. On the other side, such approaches will be also efficient in the case of multiprocessor software realization.