A Deep Dehazing Algorithm Based On GPU Acceleration

In the society,the existence of haze has had a major impact on various areas,such as traffic order,surveillance,tourism photography,and military reconnaissance.Image dehazing has played a huge role in improving people's visual senses by digital image processing technology.In recent years,the technology of image dehazing has developed rapidly.With the rapid advancement of algorithm and hardware equipment,It has become possible to realize the application of image dehazing in daily life.There are two issues that need to be addressed to accompish this goal:the one is to achieve good dehazing effect in different haze environments,The other is to realize real-time video dehazing by existing hardware equipment.In order to solve these two problems,the main work of this paper revolves around the following three aspects.(1)Targeted improvement for different concentrations of haze images.The author calculates different dehazing transmittance by edge extraction algorithm.Different weather and geographical locations lead to significant haze concentration differences in the images taken in certain haze environments.This difference requires an accurate calculation to the partial haze concentration in the image.The haze concentration of the image is determined by the depth of the area scene,and estimating the depth of the area scene is order to edge extraction algorithm.Than we make transmittance compensation for different scene depths,and areas with high scene depths receive lower transmittance compensation,on the contrary,areas with low scene depths receive more transmittance compensation.In this way,the overall sharpness of the image after dehazing is greatly improved.(2)Analysis and solve edge blurring problem of dehazing algorithm.The author refines the edge information of dehazing image by using double guided filtering technology.The first guided filtering is performed on the transmittance image to obtain a preliminary dehazing image,and then the preliminary dehazing image is used as the guiding image for the second guiding filtering,and the refined transmission is obtained by the secondary guiding filtering.Finally,the author uses this transmittance image performs the final defogging image.The edge information of the image can be utilized by the double guided filtering,so that the edge details of the filtered transmittance image is clearer.(3)The design of multi-threaded algorithm.Based on the multi-threaded GPU platform,the algorithm is redesigned to reduce latency and achieve high frame rate of high-resolution image.Each sub-step of the whole algorithm is re-scaled for large-scale parallel computing based on the GPU-accelerated platform.The GPU's high concurrency and low-latency characteristics greatly improve the overall operating efficiency of the algorithm.The algorithm achieves more than thirty times acceleration capability compared with the traditional CPU programming.The transmittance compensation method is used to make the far and near areas of the dehazing image consistent,and the overall sharpness of the dehazing image is improved by the double guiding filtering method,which greatly improves the dehazing effect of the image.Finally,based on GPU multi-threaded platform,a real-time dehazing algorithm is designed.