Researches On Visual Saliency Computation And Its Applications

With the development of Internet and the rapid increasing of image data, we hoped that the computer can process the image data as fast and accurately as the human visual system, and quickly select the effective information from it. Based on the above requirements, the model of visual saliency detection has emerged as the times require. Visual saliency detection model through mathematical model simulate and model the primary stage of the human visual system to process the image data quickly and effectively, and the salient regions are extracted rapidly from the image for obtaining the effective information. Visual saliency detection technology has been widely used in the field of image processing. It provides effective reference information for image segmentation, image retrieval, content aware image editing, the core area location, the region of interest and so on. It also helps to ease the gap between the image of the low-level features and image content understanding. Therefore, it is very important to study the technology of visual saliency detection.In this paper, we study a great number of Chinese and English literature and analyzes the deficiency of the existing visual saliency detection model. Combining with the mechanism of human visual system, we put forward two new visual saliency detection methods which are applied to the localization of the core area, the extraction of the region of interest and so on. The main research contents and innovations of this paper are as follows:(1) We describes the development of visual saliency detection model in detail and have made a summary of the visual features, the saliency detection of the law, testing standards and so on. We have made a deep research on the existing model of saliency detection and summarize the shortcomings of existing models. These has laid a good foundation for further research on the visual saliency detection method.(2) From the perspective of visual feature selection, we propose a center-focus global contrast method which is based on the contrast principle and gestalt principle. The contrast principle reflects the global color scarcity of salient region. The gestalt principle reflects the compact spatial distribution of salient region. In this method, we calculate the global contrast feature map, and then calculate the center-focus feature map. Finally, the global contrast feature map and the center-focus feature map are linear integrated to get the saliency map. The simulation experiment is carried out on the public test dataset, and the result of the simulation proves the effectiveness of the proposed method.(3) In this paper, we analyzes the shortcomings of center-focus global contrast method and proposes a saliency detection method based on multi-feature fusion. In the method, we first uses the SLIC method to segment the input image, and then extract the 10 dimensional visual features from super-pixels. And, we use BP neural network learning the relationship between visual features and visual saliency for estimating the proportion of salient pixels in the super-pixels. Finally, the salient map are generating by the proportion. Saliency detection based on multi-feature fusion is tested in a public database and compared with the existing saliency detection algorithm. Experimental results show that the proposed method is superior to the comparison of several saliency detection methods in subjective visual evaluation and objective quantitative indicators. Compared with center-focus global contrast method, the proposed saliency detection method can better adapt to the visual saliency evaluation problem.(4) The visual saliency detection method proposed in this paper is applied to the core area location and interested target extraction. Through these applications, it is demonstrated that the CF and BN visual saliency detection method is universal and practical.