UNDERWATER IMAGE QUALITY ENHANCEMENT MODEL USING ALPHA MATTE DISPARITY CHROMATIC TRANSFORMATION BASED KUAN ADAPTIVE FILTER

Image enhancement is a commonly employed technique to increase the image quality. Attaining a normal underwater image is a tedious process. The underwater image suffers from various metrics such as light scattering, attenuation, luminance problems which causes the image fogging i.e. image blurring. There is more research proceeding for the enhancement of underwater image quality, but images get blurred because of a poor visibility environment. In order to improve the image enhancement, a novel technique called Alpha Matte Disparity Chromatic Transformation Based Kuan Adaptive Filtering (AMDCT-KAF) is introduced. Initially, the underwater images are taken from the image dataset for enhancing the image quality based on different factors such as dark channel prior, salience map and chromatic map. After that, the dark channel prior is estimated depending upon the alpha matte improved transmission map to accurately extract the forefront from an input image. Then, the salience map is estimated using a disparity approach which exhibits that the quality of the individual pixels are measured by the sum of squared distance and the absolute difference between pixels. The Von Kries adaptive transformation is applied for estimating the chromatic map to control the saturation gain in an image. Finally, the resultant images are fused and then Kuan adaptive Filtering technique is applied to obtain the defog images by reducing the noise level as well as improving visibility. An experimental evaluation is carried out using an OceanDark dataset and performing the qualitative metrics and quantitative metrics such as PSNR, Absolute mean brightness error, computation time and computation overhead based on a number of underwater images and image size. The discussed outcome exhibited that the presented AMDCT-KAF model raises the image quality with higher PSNR, and minimum Absolute mean brightness error, computation time, and computation overhead than the earlier models.