INFORMATION SET BASED IMAGE PROCESSING

Abstract

Image processing plays a vital role in a variety of applications such as multimedia communication, remote sensing, astronomy, medical imaging, satellite imaging, forensic sciences, industrial automation, surveillance etc. Image processing tasks like enhancement, compression, image restoration, edge detection, segmentation and recognition are of immense use in a plenty of applications. newlineIn this thesis, some extensive image processing problems have been addressed and these are handled using Information Set theory which expands the scope of fuzzy sets by representing the uncertainty present in the attribute/property values using Hanman-Anirban entropy function. In fuzzy sets, each element of the set is a pair consisting of information source value and its membership value whereas each element is an information value in the information sets. Information set theory is applied to solve the image processing problems specifically underexposed and overexposed image enhancement, edge detection and noise reduction. These problems are covered in four chapters. newlineFirst, the underexposed image enhancement technique is introduced followed by the overexposed image enhancement. The Saturation and Value (intensity) components of HSV color model are selected to highlight the important image features by reducing the uncertainty in the distribution of image intensities using information set. For the underexposed images, keeping H and S color components preserved, V component is fuzzified. For the overexposed images, S and V components are fuzzified, keeping H preserved. New measures based on information sets are proposed for contrast information, quality factor and visual factor that give an extent to which enhancement has been achieved. Objective function is formulated based on these measures which is optimized using particle swarm optimization to learn the unknown parameters of membership functions used for fuzzification. The results of enhancement are compared with other existing methods. The proposed approach is capable of enhancing both global and local contrast and brightness as well as preserves the color consistency. newlineNext, a new and efficient framework for edge detection is proposed. This framework is originated from the smallest univalue segment assimilating nucleus, wherein a circular mask of 37 pixels is placed on each pixel of the color image for calculating a small area of neighboring pixels with similar brightness to the center pixels. A symmetric Gaussian membership function (MF) is used to fuzzify the histogram of this area. This MF is converted into sigmoidal MF to strengthen and sharpen the weak edges. These two MFs provide the best results in comparison to other MFs used in literature. An edge strength measure using information sets is proposed which is optimized to locate the robust edges. newlineFinally, noise reduction problem is discussed. A noise adaptive information set based switching median filter is proposed for denoising the images corrupted by salt-and-pepper impulse noise. This filter is an improved version of fuzzy based switching median filter which utilizes the local effective information surrounding the noisy pixel. The filtering process involves identification of noisy pixels and their removal depending on an adaptive switching criterion. This filter is capable of dealing with both low and high noise densities. Also it can preserve image details better than the existing filters. The experimental results show that the proposed information set based technique yields encouraging results in terms of qualitative and quantitative analysis as compared to some existing techniques. newline newline