Target Detection In Underwater Image

In recent years, modern technology in the field of underwater computer vision has developed rapidly. However, the videos and images shoot under water always suffer the problems of blur, color cast and noise interference. It is still an open problem of performing target detection in such complex scenario.This thesis aims to explore the principle of optical imaging system in underwater environment, and to provide an effective and robust algorithm framework of target detection. The major contributions of this thesis are listed as follows.We propose a color cast correction algorithm based on color-space translation model. Compared with other color cast correction algorithms, such as the famous Gray-World method, our algorithm provides moderate contrast improvement and avoids over-enhancement.Inspired by a simplified underwater image formation model, we propose a new image restoration model based on quaternion filtering. This method is helpful to recover the contours of target with higher quality.Since the performance of our image restoration model is most related to the selection of parameter values, we establish a parameter estimation algorithm based on a new image blur index model of quaternion phase congruency.In the whole target detection framework, the color cast correction algorithm and image restoration model are used as pre-processing modules to improve underwater image quality, which are followed by a shape matching module. The experiment results show that these pre-processing modules are helpful to improve the accuracy of target detection and location.