Research Of Image Dehazing And Hardware Implementation

As the development of image processing method and processors, the image and video processing technique has been developed rapidly over the years. And as the development of integration levels of VLSI, the ppi of sensors and monitors is increasing fast. Higher processing speed, higher resolution and higher compression ration is the tendency of all our image and video processing method. Images of high quality bring us much convenience in our work and daily life. Human vision is the most important information medium for us. The images of high quality play an important role in information science, traffic, medicine, military and et al.In environment of haze or smoke, the light of scenes is scattered by particles in the air. The degraded image becomes pale, lack of colors and image details. In the condition of outside monitoring, the degraded images will increase the difficulties of field identification. So image enhancement is indispensable for outside monitoring.Our work is based on the dark channel prior. We analysis the principles of the dark channel prior, as well as the defects of it. Then we propose three improved methods for single image dehazing based on the dark channel prior. For the first one, we calculate the coarse dark channel and the lowest channel of each pixel. Then we use the Joint Bilateral Filter to estimate the fined dark channel. For the second one, we use the Guided Filter to refine the coarse dark channel. For the last one, we choose to refine locally instead of refining the whole image. Compared to the original method, our proposed methods require less computing time.For real-time implementation of FPGA, we merge the dehazing module into the ISP. The dehazing module contains mainly about three submodules:submodule of coarse dark channel estimation, submodule of airlight estimation and submodule of coarse dark channel refining. Experiments show that the output of the dehazing module is nearly the same as the output of the Matlab simulation.