Underwater Image Enhancement Algorithm Based On Convolutional Neural Networks

Underwater small objects detection and recognition is the key to intelligent operation of underwater fishing robot.However,poor visibility of underwater images that caused by complex underwater environments results in a difficult task for underwater objects detection and recognition.Underwater image degradation contains color casts,blurring and low contrast that caused by light absorption,forward scattering and backward scattering respectively.To raise the accuracy rate of underwater objects detection and recognition,a multi-step underwater image enhancement method based on Convolutional Neural Networks is proposed.The main work of this paper are as follows:Firstly,an adaptive color correction method is applied in order to compensate the color casts of underwater images.Compared with other color correction methods,the proposed color correction method can produce more natural color corrected images.Secondly,underwater image de-blurring based on scene depth Convolutional Neural Network model is carried out.A scene depth map will be derived from color corrected image with scene depth model.Converting the scene depth map into transmission map,the de-blurred images will be acquired when applying atmospheric scattering model.This step improves the contrast of color corrected image while removing the blurriness.Finally,underwater image detail enhancement based on super-resolution Convolutional Neural Network model is applied to enhance the detail of the de-blurred images.This step improves the detail in the aspect of image resolution.Experimental results show that the proposed method improves color casts and visibility of underwater images efficiently.Compared with other underwater image enhancement method,the proposed method achieves better results.In addition,the proposed method also arrives at good results in underwater image feature point detection and underwater objects detection and recognition.