Image Enhancement Alogrithm Based On Nonlinear Diffusion Filtering And Its Implementation On DSP

In the family of image contrast enhancement algorithms, a large set of algorithms is based on the structure of illumination-reflectance decomposition. This type of algorithms decomposes an image into a low-frequency component called illumination image, and a high-frequency component called reflectance image. The enhancement effect is achieved by compressing the dynamic range of illumination and increasing the contribution of reflectance. This paper conducts research into the contrast enhancement algorithm that is based on illumination-reflectance decomposition. Several tasks are addressed, including the algorithm improvement and its implementation on a hardware system.In respect of the algorithm improvement task, we first analyze the problems, which include halo artifacts and contrast loss, in the enhancement algorithm with illumination-reflectance decomposition. We indicate that the problems mainly originate from the defect of the illumination estimation procedure of the algorithm. Then, a modified method that is based on nonlinear diffusion filtering is proposed for the illumination estimation step. In order to enlarge the diffusivity in texture areas of the image, texture suppression process is incorporated into the adopted nonlinear diffusion equation. The proposed method has two merits:first, the boundary areas are preserved in the illumination, and thus, halo artifacts are prevented; and second, textural details are smoothed away from the illumination to not suffer from illumination compression. Furthermore, we integrate several useful image enhancement techniques into the enhancement algorithms based on the illumination-reflectance decomposition. These techniques include histogram clipping and dynamic range stretching. Two acceleration schemes for the algorithm are also introduced, including AOS numeral scheme and multi-resolution technique. Several experiments in the paper show that the proposed enhancement algorithm by nonlinear diffusion filtering has better performance in both global brightness promotion and local contrast enhancement than its recent competitors.In respect of hardware implementation task, the proposed contrast enhancement algorithm with nonlinear diffusion filtering is transplanted to C674x DSP system. By means of conducting hardware debugging under CCS development software, we obtained the processed results that are consistent with those processed in MATLAB software in PC platform. In order to estimate the real-time performance of the algorithm code for surveillance application, we perform in-depth code optimization which includes software pipeline, functional resource balancing, single instruction multiple data, arithmetic operation optimization and internal memory deployment. These optimization arrangements enable the proposed contrast enhancement algorithm to obtain high operation efficiency in C674x DSP system. Specifically, the operation speed reaches 25fps for image size of 640×480.