Research On The Technologies Of Non-Photorealistic Rendering Based On Texture

The aim of classics photorealistic computer graphic is to pursuit stimulating real effects or scenes of the world. The main technology of computer graphic always applies physics modeling to stimulate properties, such as color, light, material quality. The results of these methods can manifest the world's real image or video segment. However, through stimulating the real world to realize the photorealistic effects is very complex and difficult work. On the other hand, traditional photorealistic rendering results can not satisfy people's requirement. Therefore, people need other methods to explore the new fields of computer graphic urgently.With the development of computer graphics, more and more people pay their attention to the technique of Non-Photorealistic Rendering (NPR). This technology combines artist painting rule and some scientist methods. We need this technique urgently to express and transfer information besides photorealistic computer graphics, such as decrease integrity and dependence of images and graphics. The physicals of objects in the images are expressed by depict randomicity and illegibility. The technology of NPR can cover lots of fields, such as stimulating nature scene or artist's painting work, enhancing image's information, creating animation, expressing art of date. In addition, the technology of NPR plays an increasingly important role in medical science, architectonic and education. The NPR methods can adopt image or scene's content; ignore unimportant fields, and let people's attention focus on the important content. The appearance of NPR research fields enriches the content of graphics.This thesis analyzes the limitations of photorealistic rendering, and introduces the characteristic of non-photorealistic rendering. The thesis also indicates the classification and the application of NPR. Based on the NPR of 2D texture and video texture scene, this thesis discusses some results in detail, such as texture transfer, pencil drawing, fluid artistic style. In addition, the paper researchs the technology of video texture, users can obtain an abstraction and infinite video scene. The main work in this thesis includes the following aspects: 1. Research texture deviation mapping based on 2D natural texture. The methods proposed are inspired by the texture deviation mapping constructed from a single texture background image, and texture synthesis can be fulfilled to produce the background image with the same size of the foreground image. First, we make some relevant improvement of existing technologies of texture synthesis, so that the efficiency and performance can be enhanced greatly. Further, based on the technology of Phong illumination model, we establish our method for obtaining artistic appearances taking the deviation mapping as the underlying basis. Therefore, the foreground image paints in this virtual surface can be illuminated to generate the final effect, and we also can obtain color results.2. Genetating NPR artistic effect of pencil drawing from 2D images. Many researchers draw their attention to using computer to simulate artistic painting course of people, pencil drawing is one of the most important methods in NPR. We give an overview of our research on NPR methods for computer-generated automatically pencil drawing from 2D images. By taking the vector field and the white noise images, we succeeded in creating pencil drawing with their tone matching that of the original images simply. Stroke directions are defined by the vector field and the technology of line integral convolution (LIC) algorithm. Experimental results show that the proposed method can be used efficiently to create both of the traditional or multicolored pencil drawing. In addition, we also apply these methods from static images to video scene.3. Research the technology of LIC rendering algorithm to stimulate fluid artistic effects. To generate vivid and flexible fluid artistic effect, the information such as moving and direction should be expressed in the static images. The value of pixel has relation with his own location and direction not only in this moment, but also in one duration time. Firstly, image structure vector field is derived from source luminance orthonormal vector sets or from the input image, and using neighbor vectors to improve the weak gradient fields. Secondly, LIC convolution length is calculated according to the local statistical characteristics of the texture image. Finally, the algorithm transfers the reference image's color characteristics to the rendering result to make it full of rich color. The results are vivid and flexible fluid artistic style.4. Study NP-video phenomena model system. In our system, firstly we apply the technology of abstraction to generate sketch effects of every frame from input video scene. The thesis mainly propose the method for generating the artistic appearances based on reducing contrast in low-contrast regions, and artificially increase contrast in higher contrast regions with gradient edges. Further, the important information of abstraction can be focused with the attention of the user. Experimental results show that our methods are effective.Secondly, we improve the modeling video texture based on natural phenomena. To remove visual discontinuities, the algorithm measures similarity of every two frames to determine a preliminary playing sequence and measures similarity of every two frames to determine a preliminary playing sequence. The algorithm also removes the abrupt change from the last frame to the first frame in the video loop, and the techniques of texture synthesis can be used to synthesis the frames after division and re-organization.