Research Of Fog Degraded Image Restoration Technology And GPU Realization

In the condition of fog and haze climatic, the image has the serious degradation phenomenon due to the scattering effect of the particles in the atmosphere. It will cause poor contrast and details information loss, which will influence both human visual sense effects and on the applications value of the image. So it is very important to restore degraded image affected by haze, which has significant theoretical significance and imperative reality need. As of now haze-removal techniques also have many defects, thus quality of the haze-removal image can still be further improved. This paper focuses on degradation mechanism of the fog imaging, and explores kernel technologies of fog-removal algorithm. The main works of this paper are as follows:(1)In order to overcome the bad effects of existing methods in bright sky area, a defogging method is proposed based on sky region recognition and modification.If bright sky areas exist in fog scene, bright areas are segmented by contrast adjustment procedure and the Otsu algorithm, and then atmospheric veil of the sky region is modified properly. Combined with the edge preserving characteristics of bi-exponential edge-preserving smoother filter, we can obtain the accurate estimation of the defogging physical model parameters.(2)Since the image after defogging has problem with low brightness and poor chroma, firstly, a more accurate estimate of the atmospheric light intensity is performed using morphological operations; secondly, a nonlinear luminance and chrominance adjustment method is proposed, which the parameters can adjust corresponding to the fog scene. Experimental results show that the restored image is more in line with post-processing needs.(3)In view of the existing methods have the common defects of insufficient scene adaptability, we have to estimate the parameters of the defogging physical model more quickly and accurately. Combined with the parameters of atmospheric veil in different scenery depths, a global atmospheric veil estimating method based on the fog density of different scenery depths is proposed, and the global atmospheric veil is further refined through Lee filter. Experimental results show that the method can not only quickly and accurately calculate the atmospheric veil, but also has robust scene adaptability.(4)According to the defogging algorithms has problem with low efficiency, which cannot meet the needs of high-definition images defogging in real-time. Based on CUDA programming framework, the defogging method based on fusion of the atmospheric veil is transplanted to the graphics processor unit(GPU) platform. Using the parallel computing ability of GPU’s special simultaneous operating mode, the algorithm operating speed is further promoted. Experiments demonstrate that the parallel implementation algorithm in this paper can restore a high-definition fog image in real time.