Current Issue Cover


摘 要
目的 符合用户视觉特性的立体图像体验质量评价方法有助于准确、客观地体现用户观看3D图像或视频时的视觉感知体验。现有的评价方法仅从图像失真、深度感知和视觉舒适度中一个维度或两个维度出发对立体图像进行评价,评价结果的准确性有待进一步提升。为了更加全面和准确地评价3D图像的视觉感知体验,提出了一种用户多维感知的3D图像体验质量评价算法。方法 该方法首先对左右图像的差异图像和融合图像提取自然场景统计参数表示失真特征;然后根据敏感区域深度变换结果,以及关键匹配点数目表示深度感知特征;接下来对显著区域提取视差表示舒适度特征;最后综合考虑图像失真、深度感知和视觉舒适度三个维度特征,采用支持向量回归得到最终的体验质量得分。结果 在LIVE数据库和Waterloo IVC数据库上的实验结果表明,所提出的方法与人们的主观感知的相关性达到了0.942和0.858。结论 该方法充分利用了立体图像的特性,所构建模型的评价结果与用户的主观体验有更好的一致性。
Research on 3D Image Experience Quality Evaluation Method of User Multi-dimensional Perceptual

Dong Tianyang,Yang Lijin(College of Computer Science and Technology,Zhejiang University of Technology)

Objective The stereoscopic image quality evaluation method conforming to user''s visual characteristics helps to accurately and objectively reflect the visual perception experience when users watch 3D images or videos. The existing evaluation methods only evaluate the stereoscopic image from one dimension or two dimensions of image distortion, depth perception and visual comfort, and the accuracy of the evaluation result needs to be further improved. In order to evaluate the visual perception experience of 3D images more comprehensively and accurately, a stereoscopic image experience quality evaluation method based on user’s multi-dimensional perception is proposed. Method The method firstly extracts natural scene statistical parameters from the difference image and the fused image of the left and right images to represent the distortion feature, and then expresses the depth perception feature according to the result of sensitive region depth transformation and the number of matched key points, and next extracts the parallax for the salient region to represent the comfort feature; Finally, the final experience quality score is obtained by support vector regression considering the three dimensional features of image distortion, depth perception and visual comfort,. Result Experimental results on the LIVE database and the Waterloo IVC database show that the proposed method has a correlation with people"s subjective perception of 0.936 and 0.854, which is better than other methods. Conclusion The method makes full use of the characteristics of the stereoscopic image, and the evaluation result of the constructed model has better consistency with the subjective experience of the user.