Research And Implementation On Enhancement Technology For Low Illumination Image

With the rapid development of computer vision technology, a large amount of image information can be captured by imaging systems and computer vision monitoring equipment. However, in some low–light conditions, such as indoors, night, uneven illumination conditions, as the unnatural illumination light is insufficient, the reflected light on target surface is weak, resulting in insufficient light entering the imaging sensor, which causes that the quality of image on these low–light conditions is seriously degenerating, and recognition of image is very low, and images contain a lot of noise, so that it is difficult to distinguish details in the image which greatly reduces the value of image.Traditional image enhancement algorithms are mostly about the spatial and frequency domains, the processing results usually can not meet the expected demand. In view of this situation, this thesis detailed analyze image processing theory and key technology under low illumination environment, and make some improvement on the existing methods. The concrete research contents are as follows:Firstly, this paper analyzes the research background and practical application value, and studies deeply into the traditional low-light image enhancement algorithms, such as histogram equalization, homomorphic enhancement, and the Retinex algorithm based on color constancy which is widely used. Retinex algorithm imitates the human visual characteristics, estimates the illumination component of low-light image of the illumination component, and obtains reflection components reflecting the essential characteristics of object by correlation calculation, in order to do enhancement processing on the image.Secondly, aiming at these shortages of Retinex algorithm, an improved low-light image enhancement framework based on variational Retinex theory is proposed. Luminance component of the input low illumination image is decomposed by variational Retinex theory, and reflection components and illumination components are obtained through multiple iterations; Illumination component correction function is introduced to adjust the contrast of the image; reflected component correction function is introduced to remove the noise and enhance the details; then the final enhanced image is obtained through automatic white balance process. Experiments show that the algorithm effectively improves the image enhancement and color distortion over other issues.Finally, based on LIP image enhancement model, an improved algorithm of illumination image enhancement by LIP model is proposed. This algorithm combines visual characteristics of the human eyes with the classic logarithmic image enhancement model. Luminance component of low illumination image is improved by logarithmic image enhancement model, and the dynamic range of the image and the sharpness of the image are adjusted by adaptively adjusting parameters. Because the processed image contains a lot of noise, three-dimensional block matching algorithm is introduced to denoise the enhanced image. Experiments show that the algorithm improves image brightness while effectively suppresses image noise.