Research On Several Issues Of Image Rendering And Display

Image is one of the most important carriers of visual information. As an information expression form which most closely connected with the public, image can influence the human knowledge of the world. As the spread of digital camera and the development of Mobile Internet technology, the number of digital images is increasing exponentially. An image with a soul is worth a thousand words. In the era of computer mediated social networking, people are willing to share images to show their work, life and emotion, thus, good image rendering and display method has become the key to express their intent. In addition, as the explosion of image information, identifying the importance from complex images helps the understanding of image. The visual importance detection technology can also serve image rendering and display. Therefore, the image rendering/display and importance detection methods have important research significance.Image rendering and display operate on single or multiple images to give appealing visual effect. Current methods are lack of dynamic effect and are limited to some applications. Important information detection of image is widely used and is one of the key issues of image redering and display.This dissertation focuses on some key issues of image redering, display and visual importance (saliency) detection, which include:image wandering display method, image saliency detection based on photo elements and selective image abstraction. Our contributions are summarized as follows:1. It presents an approach to wandering in the image scenes based on camera zoom style. The method greatly increases the dynamic effect and enjoyment of image display. It switches between images scenes in a continuous zoom-like process. According to visual importance analysis, the criteria to measure how well an image can be used as a patch to be embedded into another are defined, based on which a dynamic programming algorithm is proposed to generate the optimal playing sequence. We propose an effective composition method to embed the next image to the previous one to produce visual coherent output. The method has good versatility.2. It proposes a novel method for image saliency detection based on photo elements like color, depth of field and composition. This method can convey the photographers’s idea of a scene. It defines the relative importance between colors, which improves the accuracy of color saliency. Because traditional methods work poorly on low depth of field images, a simple but efficient classifier is trained to improve the detection results of such images. The detection result will be father refined by an image composition and structure analysis step. This method can serve a variety of applications in image rendering and display fields like photo collage generation.3. A selective image stylization method based on nonlinear diffusion is presented. The proposed method can generate selective abstraction results efficiently: simplifies unnecessary information and emphasizes the importance, ahieves the harmonious mixture of reality and stylization. It combines PDE (Partial Differential Equation) method with bilateral filtering and proposes an improved nonlinear diffusion model for abstraction, which simplifies inside regions as well as region boundaries. The model can control the degree of abstraction based on importance map and generate levels of abstraction effect. The model adopts the strategy which combines the isotropic filtering with anisotropic one. This method is faster than the previous models and can produce more sophisticated abstraction results.