Novel Approach to Development of Direct Volume Rendering Algorithms Based on Visualization Quality Assessment

摘要

A novel approach to the development of direct volume rendering algorithms is proposed. The approach is based on the assessment of the visualization quality. Analysis of rendering artifacts and various assessment methods is done using 3D reconstructed medical tomograms as test datasets. Analysis of the previous research on methods of 3D visualization quality assessment and quality improvement is presented. 3D visualization artifacts and quality measurement method (quantitative estimation) is proposed. The method does not require any ground truth image; hence, it is a universal approach to measure the quality of the 3D-reconstruction by any ray casting technique. The method is based on generation of the reference image as a mathematical expectation for a set of 3D visualizations obtained via the jittering technique. Start position of the rays are being shifted randomly towards their directions. To estimate the deviation of a pixel value from mathematical expectation, we use PSNR metrics, which is traditional metrics in signal and image processing and measures deviation in the decibel scale. The conditions of a proper use of the technique are discussed. A new virtual samplings method with preintegration is proposed to reduce sampling artifacts in the ray casting algorithm. The novelty of the method consists in using pre-integrated classification instead of post-classification in the virtual sampling method. A novel approach to 3D visualization algorithms development based on analysis of a ray casting method in quality-performance space is demonstrated by comparing the state-of-the-art ray casting methods and the proposed method. The comparative analysis revealed an advantage of the classical pre-integrated method in the case of using trilinear filtering without local shading. It also showed the advantage of the proposed virtual sampling method with pre-integration in the case of using tricubic filtering with local lighting.