Studies On Image Assessment And Video Coding Based On Visual Attention Model

Visual attention scheme existed in human visual system is the scheme to lead our sight to notice the interesting object in the scene. It allows human visual system to get more useful information from the limited system resources. Visual attention model has been one of the hot research topics in image processing. This paper focuses on the research work about visual attention model. First, bottom-up visual attention model is studied and then a new visual attention model integrated with image location information is presented. Secondly, a new image assessment metric based on visual attention model is proposed. Then, a moving objects detection algorithm based on brightness distortion and chromaticity distortion is proposed. Last, visual attention model and moving objects detection algorithm are used into H.264 encoder.In existed visual attention model, color, texture and orientation are used to calculate the saliency of each pixel. At the same time, human has different sensitivities for different locations in a image. In this thesis, we use the location information of each pixel to calculate its saliency and get the saliency map of the image. Experimental results show that the visual attention model integrated with image location can fit HVS better. In image assessment area, traditional PSNR is the mostly used image assessment metric. But in most cases, high PSNR dose not mean good image quality. In this thesis, we use saliency map to generate a new image assessment metric based on perceptual interest. Experimental results show that the new metric accords with HVS better than PSNR for the images of different quality.Moving objects detection is a key technique related with video coding. For example, the area of moving objects are"the most interesting"region in smart video surveillance and need higher coding quality. A new moving objects detection algorithm based on brightness distortion and chromaticity distortion is proposed. Statistical analysis of the chromaticity value is used in background extraction. It can solve the problem of the remaining traces of moving objects in the case of few frames and slow movement. Gaussian model is used to calculate the threshold of brightness distortion used in object detection. It can solve the problem of manually setting the brightness distortions of different video sequences. Experimental results show that compared with other moving objects detection algorithms, our algorithm can exactly extract background frame of a video in the case of few frames and low movement. It also can adaptively calculate brightness threshold according to the characteristic of different video sequences and them detect the moving objects exactly. Then, a new video coding method based on visual attention model and moving objects detection is discussed. We aim at the optimization of subject image quality, considering the difference of the sensitivities of ROI and Non-ROI, propose a new rate control algorithm based on visual attention model. In all of the rate control schemes, to set the quantization parameter of a single micro block is the most difficult problem. In this thesis, a method is proposed to set each micro block's quantization parameter based on saliency map and moving objects. Experimental results show that the coding method can give better subject quality.