Research On Enhancement And Noise Reduction Of A Low-light Image

With the rapid development of computer vision and digital image processing technology, image devices are widely used in the field of public security, intelligent transportation, industrial production and so on. A clear and high quality image can provide effective information for human or computer vision system. However images obtained under low-light conditions tend to have the characteristics of low gray levels, indistinguishable details, and high noise levels. Image degradation caused by low-light conditions not only affects the recognition of images by the human eyes but also influences the performance of computer vision systems. Therefore, enhancement and noise reduction algorithms of low-light images are proposed to improve the image quality.Firstly, typical image enhancement and de-noising algorithms are studied, and we focus on the analysis of basic principle of the low-light image enhancement algorithm based on the dark channel prior de-hazing technique. This algorithm used the de-hazing technique to process the inverted low-light video frames by comparing the similarities between inverted low-light intensity images and images with dense fog, which can significant improve the image contrast. However, the enhancement results contain obvious block artifacts, and the step of estimating dark channel map is time-consuming. A fast low-light image enhancement algorithm is proposed in this study aiming at improving the drawbacks of the traditional algorithm by analyzing the similarities between the luminance map of inverted low-light intensity images and corresponding dark channel maps. By using relatively smooth luminance map instead of dark channel map, the proposed algorithm can effectively avoid the block artifacts, improve the stereoscopic effect, and greatly reduce the processing time, which can improve the visual effect and time performance of the traditional algorithm.Secondly, by analyzing the sensitivity to noise of the enhancement algorithm based on the de-hazing technique, we improve the physical model of the inverted low-light images, and propose a simultaneous enhancement and noise reduction algorithm. By comparing the similarities between the improved inverted low-light image model and the kernel regression model, the problem of simultaneous enhancement and noise reduction is converted into the least squares problem using the framework of kernel regression. The iterative joint bilateral filter is used to correct the parameters of inverted low-light image model alternately, and the quotient image method is used to compensate for the details in each iteration step. The proposed algorithm can effectively improve the image contrast while reducing noise and preserving the details.Finally, we evaluate the performance of the typical enhancement and noise reduction algorithms and the proposed algorithm by comparison experiments. Compared with the typical enhancement and noise reduction algorithms, the proposed simultaneous enhancement and noise reduction algorithm based on the de-hazing technique performs well on the subjective visual observation and objective quality measures, and can effectively improve the image contrast while reducing noise.