Underwater image visibility improving algorithm based on HWD and DehazeNet
Panwang Pan, Fei Yuan, Guo Jia, En Cheng · 2017
It is meaningful to find an effective method to solve the problem of underwater image detail loss and contrast decrease caused by turbulence. In this work, we present a novel method to enhance edge and the contrast of the image. First, we separate the high frequency and low frequency parts of underwater image by using the HWD, and then we remove the image noise and enhance the edge of images. We also employ a trainable convolutional neural network named DehazeNet to estimate transmission map of underwater image. In order to obtain a more accurate transmission map, we apply a multiscale iterative framework and adaptive bilateral filter to filter the transmission map. Experimental results on blurred images show that the proposed algorithm can visually enhance the edge and remove the turbulence blur to achieve the effect of clarifying the image. Compared with the existing three methods, our algorithm can enhance the image naturally. And our method's color deviation is smallest, the objective quality score is highest.