SOME STUDIES ON IMAGE ENHANCEMENT AND FILTERING

Abstract

Images are one of the best sources for research & communication. Proper analysis of image data is essential for various fields like satellite images & remote sensing, Biometrics, Criminology, military surveillance, astrology & many more. Requirements of the users vary according to the nature of application in use.Image processing plays a critical role in the analysis of data and is an integral part for various applications. Image denoising is one of the challenging branches of image processing. Impulse noise is among the prevalent noises that degrade the image quality and its subsequent noise suppression plays a pivotal role in the enhancement of images. The presence of impulse noise cannot be averted during the digitization, acquisition, and transmission of images. Many state of art filters are available in the literature to deal with impulse noise encountered in images. Filters having dual modes of detection and restoration exhibit superior performance in removing noise and thereby keeping the original information of the images intact. In real world sometimes the user is uncertain about his requirements therefore the characteristics of the employed filter for impulse noise removal should be adaptable to the indecisive features of an image.The denoising filter should be robust enough to handle the varying amounts of noise density and should be intuitive in nature. This thesis work tries to cater to all the essential features required for an efficient filter. Incorporation of fuzzy logic makes the filter more versatile.Investigations performed in this thesis show that the proposed work excels in the quantitative as well as qualitative manner. Four schemes introduced in this thesis are : (i) High-density impulse noise detection using FCM algorithm (HDIND) viii (ii) Edge preserving fuzzy filter for the suppression of impulse noise in images (EFFSIN) (iii) Heuristic analysis of neighboring pixels for impulse noise detection (SPHN) (iv) Impulse noise removal in color image sequences using Fuzzy logic (INFL) The initial three schemes namely HDIND, EFFSIN, and SPHN focus on the grayscale images and INFL is proposed for color image sequences. All the schemes incorporate an efficient detection criterion and after proper classification of noisy and noise-free pixels, performs the restoration procedure. The use of fuzzy logic in the methods has enhanced the decision making aspect of the algorithms to classify the noisy pixels present in an image. The simulation results are done in isolation for all the schemes deduce that HDIND and EFFSIN are robust in nature and their performance does not deteriorates with a rise in noise density. The edge-preserving nature of EFFSIN preserves the original image data and false alarm rates are reduced. SPHN provides good PSNR and MSE results. INFL is a Spatio-temporal filter that gives excellent performance. SSIM, PSNR, MSE, False alarm and Miss detection rates are used as quality measures to analyze the proposed mechanisms.