Research On The Non-Photorealistic Rendering Of Image And Video Based On Content Analysis

During the creation process of artists, the viewpoints always release from therestriction of physical rules of the real world, to focus on the expression of subjectiveimpression, but not simply reappear the real scene. Non-photorealistic rendering (NPR)technology is trying to use computer to simulate the creation process of artists, or helpartists to create such artworks. To be specific, by observing and studying the creationprocess of artists, NPR picks some works for computer to automatically process,improving the efficiency with the computing ability of computer, to help artists focus onthose most creative work. The goal of NPR is not the photorealistic effect of images, butto express the artistic feature of images, simulating the real artworks, or as thesupplement of photorealistic images. Nowadays, NPR has been used in many areas ofindustry, and has made some achievements. Since NPR has attracted more and moreresearch interest, it contains very important meanings of research.This paper considers several problems in non-photorealistic image and videorendering technology. Based on low level feature and high level semantic informationof images and videos, this paper discusses several issues which includes the artisticimage generating technology based on edge analysis, the technology of regionclassification based on image segmentation, artistic video rendering by video pixelmotion clustering, and sketch image rendering with simulation of light. This paperpresents the automatic generating algorithm of two specific artistic images. Byanalyzing the image edges, considering the feature of human visual perception, hidingthe secondary object into host image or embedding into outer object, to achieve specificartistic effect. This paper also discusses the technology of image region classification,by using the coarse regions by current automatic segmentation algorithm as input, toclassify and select appropriate rendering style and parameters for each region. For videorendering, this paper presents the painting patch technology based on video pixelmotion clustering. By mapping every region with similar motion to one reference frameto construct panorama image, and reverse map the rendered panorama back to everyframe to composite the output video, this technology could improve the framecoherence and stroke consistency which is a common problem in previousnon-photorealistic video rendering algorithms. This paper also presents a sketch image rendering algorithm based on simulating light. By simulating the light in the scene, toachieve the light and shadow effect in sketch image, which is much correspondencewith artists’ creation style.