| Along with the development of modern information-based society, image has become a major source of human access of information. In the process of image forming, however, various factors reduce image quality, such as insufficient light, uneven illumination, bad weather. They can affect people's perception and are not conducive to the post-processing of the image. Therefore, on the basis of Retinex theory, this paper proposes an image enhancement method to improve image quality.Retinex theory is a color constancy perception theory. Based on the Retinex theory, a variety of Retinex algorithms was proposed, the most eminent one of which is the multi-scale Retinex with color restoration. Nevertheless, when the image is saturated colors, multi-scale Retinex with color restoration is at risk of inverting colors in the image. Moreover, at the gain compensation stage, we find that the small probability value at both ends of the histogram occupies too much gray level with the result that the gray level of image useful details is too small, resulting in blurred images. Therefore, this paper studies the causes of color inversion and the method of gain compensation. Then an improved multi-scale Retinex algorithm is proposed in this paper. We use tansig function instead of logarithm function. It effectively eliminates the color inversion phenomenon. At the same time, we propose a gain compensation method based on the combination of histogram stretching and Gamma correction, which can effectively improve the image details, so that the image contrast is better.This paper selects a variety of different types of degraded images, such as bad weather image, low illumination image, underwater image, inhomogeneous light image, color cast image. Then this paper compares the algorithm with the different algorithms domestic and abroad. From two aspects of subjective evaluation and objective evaluation, we show that the image enhancement algorithm proposed by this paper is universal and effective. |
PDF Full Text Request