Some Studies on Underwater Image Enhancement

Abstract

Oceans covers 70% of the earth and are required to be explored for various scientific newlinestudies like setting up underwater pipelines, knowing the ecological balance by taking population newlinecensus of various aquatic species etc. To get such information, capturing images is newlinesafe and effective solution, but underwater imagery suffers from poor contrast, bluish green newlinecolor cast, hazy appearance, blurring and noise due to inherent property of water. newlineEnhancement of underwater images is a challenging task as images are captured in different newlinemedia having varying salinity, turbidity, green matter etc. and captured at varying depth. newlineThus, there is a need of an adaptive method which works for all type of underwater images. newlineThis thesis provides an effective solution to pre-process the underwater images to make newlinethem suitable for scientific studies. Under the umbrella of pre-processing, sub-problems newlinelike color correction, contrast correction , haze removal have been addressed irrespective newlineof the underwater media and depth at which the images have been captured. The contributions newlinepresented in this thesis are outlined below: newline We propose an adaptive method for enhancement of underwater images by using newlinefuzzy rules to find the color cast. Since the color correction depends upon the media, newlinesome parameters required for controlling the degree of color correction are found newlineusing Multi-Objective optimization (MOPSO). Performance measures which are optimized newlineusing MOPSO act as proper guiding mechanisms so that the resultant images newlinehave few or no artifacts. Various non-reference based performance measures like entropy, newlinehistogram spread, UICM have been used to analyze the results. newline We propose a local enhancement solution which tackles the artifacts caused by artificial newlinelighting. It performs color and contrast correction in a localized manner. The two newlineversions are then fused using wavelet based fusion with the help of effective weight newlinemaps which chooses the required features from the two versions. newline