Research On Haze Removal And Image Decoding Accelerating

Image degradation is a key problem when making most use of imagingtechnology. Generally, sunlight transmit though the air to the object, the light reflectedby the object run into the imaging system. Because of the existing of small droplets infoggy day, the light intensity get into the imaging system degraded. Meanwhile, thesesmall droplets also reflect light to imaging system from sun. The mixture of this effectby droplets degrades images from clear sharp ones. Based on the phenomenon ofabove mentioned and the math model of imaging, this thesis proposes an adaptivehaze removal method using dark channel prior, moreover revaluating the air light.Through the analysis of a large number of experiments on foggy images, the novelproposal get better results in dark light and multi light source. The result image isclose to real image.Images in foggy days tend to have higher value. Large numerical floating pointoperations require more time. Through the comparison of pixels in with-fog andwithout-fog images in same position, the result of IDCT(Inverse Discrete CosineTransform) running time reveal the with-fog images spend more time than thewithout-fog images on transforming. Based the conclusion, this thesis propose adesign of JPEG decoding accelerating based on OPENCL. IDCT, Inverse quantum,color space conversion, up sample will be paralleled by the GPU which own moreefficient computing performance. The result turns out, GPU based JPEG decoding runfaster than C version, offer an efficient software solution.