Order Statistics - Fuzzy Approach in Processing of Multichannel Images and Video Sequences

Abstract

There exist different applications of the image processing, such as medical imaging, high definition television, virtual reality, remote sensing, ultrasound and radar imaging, etc. In these applications, it is necessary to restore an image (or frames of video sequence) and decrease a noise influence exploiting the filtering algorithms that form a part of a general image processing system. The images are corrupted by noise, in sensors employed or maybe, during signal transmission. Also, several kinds of noises are produced by natural phenomena (atmospheric, scattering, interference, etc.). Usually, real noises are described by different models, there exist impulsive, additive and multiplicative (speckle) ones. So, the image pre-processing efficient scheme should be one of important part in any vision application permitting to suppress a noise, saving the image performances, such as, edge and fine features preservation, and also the chromaticity properties for the multichannel (multispectral) images. This demands to have several efficient filtering schemes, which depend on noise type and priory information, in a pre-processing stage of image or video sequence processing system. The main objective of present chapter is to exhibit several justified approaches in restoration of the images and video sequences, which are usually affected by noise of different nature, which can be efficiently used in different applications of the multichannel (multispectral) images and sequences. Here, multispectral image is defined in such a way, where each a pixel is represented by a number of channels that carry out information about its spectral content. Multispectral images span the domain of images from conventional three-channel colour images to hyperspectral imagery with hundred of bands/channels used in remote sensing applications, medicine, spectrometry, etc. In literature, there exist a lot of algorithms that process two dimensional (2D) images (Franke et al., 2000); (Russo & Ramponi, 1996); (Schulte et al., 2006, 2007a, 2007b, 2007c); (Shaomin & Lucke, 1994); (Nie & Barner, 2006); (Morillas et al., 2005, 2006, 2008a, 2008b, 2009); (Camarena et al., 2008, 2010); (Ma et al., 2007); (Amer & Schroder, 1996); (Xu, 2009). We compare the proposed 2D fuzzy framework with recently presented 2D-INR filter based on fuzzy logic (Schulte et al., 2007b), where a noise is detected preserving the fine features and edges in an image. Also, other promising classes of 2D processing algorithms are employed as comparative ones: 2D-AMNF, 2D-AMNF2 (Adaptive Multichannel Filters)(Plataniotis & Androutsos et al., 1997); (Plataniotis & Venetsanopoulos, 2000); 2D7