Context aware image enhancement for online fish behaviour monitoring

Abstract

Image enhancement is an essential step in developing a robust computer vision system. This study presents an image contrast enhancement method for an online fish behaviour monitoring system. The method executes in three steps. First, the critical information on the context of the image is derived from different responses of two edge detectors: one with high threshold and the other with low threshold, which differentiate the image structure into the good contrast part and the poor contrast part. Then, statistics of the gradient field in the former part are used to modify the gradient field in the latter part. Finally, in the modified region, the image contents are reconstructed through solving a Poisson equation with the Dirichlet boundary condition. Comparison with state-of-the-art methods has been made both qualitatively and quantitatively using realistic images and advantages of the proposed method have been demonstrated.