Nonlocal Variational Image Enhancement Based On Retinex Theory

Image enhancement, as one important part of digital image processing research, is to improve quality of image. In other words, images after image enhancement processing are more suitable for human vision system or machine to analyze and recognize. And now image enhancement processing technology has been widely used in medical diagnosis, fingerprint identification, intelligent transportation, satellite image processing and other fields.Recently, Retinex theory well explains characteristics of human vision system, like color constancy. Then the image enhancement variation models based on it have been widely used in the field of digital image processing. These models and its algorithms are simple and fast. What’s more, the enhanced image is clear and vivid.In order to improve image quality after enhancement, these models should be further improved. Especially for color texture image processing, image characteristics, such as texture, edge, smoothness, should be fully preserved in image processing.In this paper, on the one hand, we apply with nonlocal differential operator based on patch-distances to some existing variation models of Retinex, which make full use of similarity between pixel’s patch. We also introduce Split Bregman algorithms of these models. On the other hand, the double TV regularized model based on Retinex theory (DTV) and its Split Bregman algorithm has been proposed after deeply analyzing existing Retinex variational models as well as its advantages and disadvantages. And then the nonlocal differential operator is applied in these model, namely the nonlocal double TV regularized model (NL_DTV). Experiments show that the DTV model and Split Bregman algorithm presented in this paper have higher computational efficiency and accuracy. It is much easier to find out appropriate parameters of these models. Compared with other nonlocal variation models of Retinex theory, the illumination L recovered by NL_DTV is much natural and smooth, in which we can barely see any detail. And shadow of original images of NL_DTV is almost removed.