Research On Single Image Dehazing

With the deterioration of global atmospheric pollution, frequent fog/haze weather influences people’s daily life more and more seriously. Due to the atmosphere particles that absorb and scatter the light, the images of outdoor scene taken in such weather conditions are characterized by reduced contrast and faded colors. Restoration of such degraded images has caught increasing attention in the last years. This task is important in several outdoor applications.In this thesis a novel single image dehazing algorithm based on fast multi-scale Retinex is proposed. Considering the correlation between the haze image degradation model and the Retinex theory, it is proved that the transmission can be estimated with the multi-scale Retinex algorithm. The new calculation method of transmission costs less time and memory space than previous methods.To reduce the computing time of the Retinex, the concept of white light channel is introduced.Instead of three color channels, the proposed algorithm processes only the white light channel.Additionally the bilateral filter which can maintain more edge information is used to replace the Gaussian filter in the Retinex processing. And then the filtering result is normalized and regarded as the estimation of transmission. The experimental results show that the proposed algorithm is quite fast and effective.