The Research On Retinex Theory And The Application In Image Enhancement In Compressed Domain

Recently Retinex has attract more and more attention. As an image enhancement method, the aim of Retinex is to resume the lost image detail caused by the limitation of acquisition equipment. Retinex method is theoretically based on the color constancy of human visual system. The basic strategy is simulating the process of human vision to enhance image contrast and resume true color of object.In this thesis, Retinex theory is the major research point, it presents a novel Retinex algorithm, based on the Retinex model analysis, and sets Retinex theory in JPEG framework, to enhance the images in compressed domain. The main content of this thesis is shown as follow:1. This thesis researches the Retinex theory deeply, and points out the feature and limitation of the Retinex based on the analysis of the mathematic formula, and then proves the conclusion in experiments. Based on those researches, it analyses four classical Retinex algorithm: random-walk Retinex, Frakle-McCann Retinex, MSR and RSR, by comparing the strategies of choosing the paths of selecting pixels, at last it offer a new thought of researching Retinex.2. A new Retinex algorithm using IIRlow-pass filter is presented based on the research on the limitation of existed Retinex algorithms. This algorithm uses 2-D path to compare the selected pixel, with linear iteration calculation method to reduce the complexity; records the edge imformation through the process of illumination estimation, to prevent the pixels with high contrast value to affect each other, this can compress the"halo effect". Besides, aimed to the limitation of existed Retinex algorithms that cannot select parameters adaptively, it combines the IIR low-pass filter Retinex with genetic algorithm which is coded by real, and adjusts the parameters in CIELab color space. The experiments in different light condition and different imaging condition show that it can pick up the suitable parameters to enhance the images, and the results are with clear details and great color.3. In the basic of recently research for image enhancement algorithm in the compressed domain,we joining the Retinex theory to the JPEG code process can achieve the goal for enhancement the JPEG compression image. The procedure may divide into two steps. First, uses the Retinex algorithm to carry on the enhancement and to improve the Luminance quantification table .It take the Retinex theory as the foundation,regarding the DC coefficient in DCT coefficient as the incident light component and the AC coefficient in DCT coefficient as the reflected light component.the goal for revises the DC coefficient is to adjusts the scene illumination condition and for the AC coefficient is to enhances the image contrast gradient ,then judges that if the the great block are activite and to revision the luminance quantification table adaptively to achieve the goal to preserve enhancement detail and suppresses the massive effect.