Optical fish estimation and detection in noisy environment

Detecting fish in submarine environment is a challenge due to the properties of the water such as light absorption and scattering. In this work, we present a method for preprocessing images in submarine environment. In the first step, we model the underwater environment as overlapp of two processes. The first process is considered as a Poisson distribution, while the second one is considered as a Gaussian mixture. The resulting distribution is called Poisson-Gaussian mixture (PGM). To estimate the noise parameters, we propose an iterative algorithm based on the expectation maximization approach. This allows us to jointly estimate the scale of the Poisson parameter as well as the standard deviation and the mean of all Gaussian distributions. In order to facilitate the detection of objects, to correct the illumination problem of the scene and to restore the colors, we integrate a color correction algorithm. Finally, detection and localization of fish complete the pre-processing in the images. To obtain medium or small regions, the mean shift algorithm is used with a reduced threshold. In the segmentation process, the proposed detector scan the image region by region. This detector allows to estimate statistically the type of the region (object or non-object). The method is tested under different underwater conditions. Experimental results show that the proposed approach outperforms conventional methods.