Research For Non-photorealistic Rendering Based On Multiresolution

Computer art is a new cross-disciplinary combining sciences and arts. It is a new mean of artistic creation to show the fresh artistic thought and artistic works. In the recent years, non-photorealistic rendering (NPR) technology becomes one of the most active branches in computer art. Non-photorealistic rendering highlights greatly in subjective sense, mood as well as painting methods, which are giving the better enjoyment for us, through its rich artistic expressions. The rendering systems and processes based on multiresolution are in line with human vision and ways of thinking. In order to represent the true painting processes and results, users may design resolution brushes with different sizes and shapes and so on to simulate the trend of lines and color brightness. The combination of multiresolution with the classic non- photorealistic rendering methods can improve the implementation effects of the original rendering algorithms and the interactive algorithms for the speed and convergence properties.Closely surrounding the subject of non-photorealistic based multiresolution, this dissertation first generally speaks of the basic approaches in NPR, stroke-based rendering, non-photorealistic rendering based on fluid simulation and non-photorealistic rendering based on texture synthesis. Then the dissertation has intensively studied the connections among wavelet transform, subdivision, reserving subdivision and multiresolution. Based on the above research, three new methods are presented in this dissertation. The main contributions of this dissertation are summarized as follows:Firstly, an approach combining multiresolution with statistics is proposed in the dissertation. Gaussian pyramid is used to generate three reference images with different resolution by according the input image. The approach paints layer by layer from the lowest resolution to the highest. Rendering each layer, the approach uses statistics method to get the brush color which is the largest number of pixels having. Constructing stroke lists, all parameters are set up to produce the painting brush. The experimental results show that the approach has got more richly stroke color, achieved a finer and smoother image.Secondly, a method based on stochastic brush is presented to create oil painting in the dissertation. Considering that there is much randomness in the painting process, the method inputs the original image and gets plenty of numbers which is implemented in VC++ to generate a random image. Then beginning with a random point, the algorithm adopts pixel'eight- neighbors concept to generate the stroke lists, and simulates the original image by setting up different parameters. The experiment results show that the algorithm efficiently and successfully simulates the randomness in the process of painting.Thirdly, an algorithm combined mutiresolution with texture synthesis is presented to create comic. After sampling circinate lines from the original image, the algorithm uses quilting synthesis technology of multiresolution to generate texture reference image. A painterly rendering is drawn with improved random brush strokes, according to reference image gradients. A natural and more influential artistic image is generated by setting up different parameters.