Clustering Based Approach forUnderwater Sunlight Flicker Removal

Abstract

Computer vision-based approaches have been fre-quently deployed for AUV navigation, ecological surveys andstructural analysis of the seafloor. However, It undergoes sig-nificant distortions due to the specific propagation property oflight in underwater environments. It includes colour imbalance,non-uniform illuminance, low contrast and blurring effects. Apartfrom these limitations, sunlight flickers (caustic noises) are one ofthe prominent artefacts found on seafloor images in shallow waterenvironments. This noise decreases the efficiency of algorithmswhich exploit velocity estimation and image feature extraction,such as navigation and mapping algorithms for AUVs. Therefore,the removal of such noises becomes essential for better missionplanning and 3D Reconstruction of sea floor. In this paper, wepresent a novel online and computationally efficient algorithm forAUVs to minimize the sunlight flicker distortions in the framesduring the video survey. The primary benefit of the algorithmslies in the fact that it does not affect the non-flicker region. Hence,preserving the local features of the raw image and maintainingthe sharpness