Research And Application For Image Enhancement Based On Guided Filter

With the popular of consumer and professional digital products, digital image processing is closed to our daily life more and more. Digital signal processing technology is applied to image enhancement can improve the vision, which can avoid some trouble such as sensor, shaking, light, rain or snow. But research on robust and efficient algorithm is difficult. Contrast stretching based on probability statistics and spatial domain convolution based on differential mask can both sharpen the image with some noisy. Image enhancement in frequency domain always restrain low frequency information, which can lose the details in smoothing area. A image enhancement algorithm can hardly satisfy requirements at the same time such as restraining noisy, compress dynamic range, boundry preserving and color restoring. Guided filter is a kind of local linear image filtering which can boundary-preserving and smoothing according to local gray variation. For obtaining the enhanced image, we can cumsum the difference between input and output to the original Image.Guided image filter has an advantage of low decoding cost, and it export gray value according to neighborhood imformation. It process the coloured image through RGB channel leading to the color distortion, bring out the texture missing and smoothing excessivly. So it is difficulty to apply it to face recognition. In order to solve above problems, firstly this paper produced the image enhancement theory in spatial domain and frequency domain, and analyze the bilateral filter and Retinex image enhancement. Secondly this paper combined the Retinex image enhancement algorithm with guided filter, used guided filter to aquire illumination and optimize the color restoring functiuon. According to the experiment, the proposed algorithm can avoid the halo artifacts that happened in MSRCR. Thirdly we produced the fractional order differential theory and construct the adaptive mask operator, it can bring the image local texture detail to guided filter, so we call it recombination guided filter. At last, we apply the proposed method to face recognition based on PC A and SVM. During the simulink experiment, we found the proposed method can mostly reserve the main constituent and compute as fast as guided filter. Above conclusion come out from objective value of the filtered image and recognition rate.