Design of fusion-based underwater image enhancement method in video stream

Abstract

This thesis has designed a fusion-based enhancement method for underwater images and video streams which are mainly distorted by color changes, poor illumination, and water particles. The main objective of image fusion is to achieve a comprehensive result which is appropriate for the human visual system. In this thesis, our goal is to use a straight forward and not computationally intensive reconstruction. As a result, the underwater images or video streams are partially restored from the original ones. Firstly, the 5.8 GHz transmitter and receiver set are used to carry the input data from the camera which is placed inside the underwater scene to the computer for the enhancement stage. The corresponding data which is received by the computer then becomes the input or original hazy image. The input is white-balanced to remove the distorted color and contrast-enhanced to recover the fine details. Color compensation is carried out to remove misbalanced color in a poorly illuminated underwater scene. The image resulted from these stages are then judged by three weights: global contrast weight, local contrast weight, and saliency weight. The purpose of these weights is to decide the quality of each pixel to let them contribute to the final output. The multiscale image fusion is carried out level by level of two pyramids: Laplacian pyramid where the input is the white balance and contrast-enhanced image, and Gaussian pyramid where the input is the normalization of three weights. Finally, the fused output results with more appealing color and less noise. The thesis has constructed a testbed to implement the proposed fusion-based underwater image enhancement method and successfully processed the noisy underwater scene. The research operates as a mid-range underwater monitoring system with a multi-resolution fashion which is prone to artifacts. This thesis will serve as a basis for future underwater monitoring or smart fish farming architectures.