Adaptive Tone-Preserved Image And Video Detail Enhancement

During the image and video acquisition process, some factors such as lighting, equipments and noise will cause bad visual effects. Therefore we need to employ detail enhancement algorithm to highlight the image and video details which make recognition and subsequent processing more conveniently for users.The existing image detail enhancement methods mainly exist in the following three problems:(1) For different input images, these methods must use different detail enhancement models or set different parameters according to the experience, which are too difficult to the nonprofessional.(2) The experimental results are difficult to retain the tonal distribution of the input images which make the detail enhanced results unnatural.(3) There does not exist good detail enhancement method for video application. In order to solve these problems, we further study and research the image detail enhancement techniques, and propose the corresponding solutions. The main content of the dissertation includes,(1) We study a variety of multi-scale image decomposition methods, however, in order to achieve real-time edge-preserving capability against any color images, without resorting to sub-sampling or quantization, we adopt the linear filter to achieve multi-scale decomposition based on domain transform. This method can quickly divide the input image into a base image which contains the coarse-scale image information, and the detail layers which contain the fine-scale details. (2) Although we use the edge-preserving filtering algorithm, a lot of strong edges and concussions are still remained in detail layers. As a result, if we enlarge the same size for strong edges during the process of detail enhancement, halos and gradient reversals will be caused. Therefore, in order to effectively restrain these phenomena, the paper construct an adaptive detail enhancement function based on the edge response which can increase the enhanced amplitude of small details while reduce the magnification of strong edges.(3) An effective image detail enhancement algorithm must guarantee that the tonal distribution of the processed image will not bring in distortion while enhancing the image details. In the process of detail enhancement, the whole luminance channel of the image still has dramatic changes even using our detail enhancement function. This shortcoming will make the tonal distribution of the synthesis image drift too far from the input image. Neither image contrast stretching nor histogram matching can effectively correct the tonal distribution of the detail enhanced image. Consequently, based on the Markov random field theory, we put forward an energy optimization algorithm which can effectively correct the tonal distribution of the image.Our experimental results show that tone-consistent image detail enhancement effect is available for arbitrary input images with unified parameters setting, which is superior to the state-of-the-art methods. And, any edge-preserving multi-scale decomposition method is applicable in our algorithm.