The Research On The GPU-based Non-photorealistic Volume Rendering

Non-photorealistic rendering (NPR) is a computer graphics rendering technique that generates not realistic but artistic creation style pictures; and GPU-based volume rendering is another technique that implements volume rendering using parallel proc-essing mechanism of modern programmable graphics hardware. This thesis studies NPR, GPU-based volume rendering techniques and focus on the contour extraction and enhancement methods of non-photorealistic volume rendering (NPVR). The main content of this thesis can be outlined as following three aspects: the technology framework of NPVR applied in programmable GPU pipeline is analyzed and identi-fied; an improved 26-neighborhood gradient estimation algorithm is proposed; a GPU-based fast volume contours extraction algorithm is designed and implemented using the improved 26-neighborhood algorithm and maximum intensity projection (MIP) technology.Firstly, the characteristics, classification and critical methods of NPR are studied. The programming mechanism of GPU is discussed according to the architecture of modern programmable graphics hardware. After analysis of GPU-based volume ren-dering and NPVR, a scheme to implement NPR using the parallel pipeline of GPU is identified.Secondly, advantages and disadvantages of traditional image-based gradient es-timation algorithms are analyzed. Considering the high gradient values of voxels on contour lines, a 26-neighborhood gradient estimation algorithm is employed and im-proved. This algorithm constructs a 4D linear regression equation and estimates gra-dient value of voxels by square error calculation. Then, contours of volume data could be extracted based on these values. Image artifacts are efficiently reduced and the continuity of contours is enhanced in experiments.At last, a GPU-based fast volume contours extraction algorithm is proposed, in which the improved 26-neighborhood algorithm is used to detect contour lines of volume data, a window function is employed to filter gradient signals, contour ex-traction and enhancement is achieved by combining maximum intensity projection (MIP) technique and GPU programmable pipeline. High quality images and interac-tive rendering speed are shown in experiments.