网格运动和视差相联合的立体视频预测

A stereo video predictive method of joint mesh motion and disparity

提出了一种基于网格的运动和视差相联合的立体视频预测算法。利用运动估计和视差估计中要保持网格拓扑结构不变这一前提条件,由当前左图像及其网格,对当前右图像进行前向视差估计;然后利用形变后的网格和上一帧右图像,对当前右图像进行后向运动估计;以三角形为预测单元,选择这两种预测中使得当前右图像中三角形的预测误差较小的作为该三角形的最终预测,从而提高预测图像的质量。最后给出了该算法的一个简化算法:连续网格形变的运动和视差相联合的立体视频预测算法。实验结果表明,该方法及其简化算法融入到立体视频编码系统后,在大致相同的压缩比下所得到的解码图像的PSNR分别比基于网格的兼容性算法高出0.6dB和0.5dB,比MPEG-4TPS能提供更好的主观质量。

A stereo video predictive method of joint mesh motion and disparity estimation is proposed, and the foundation of the proposed method is keeping the topology structure of mesh unchanged in motion/disparity estimation. First, the disparity is estimated forwardly using the current left image and its mesh. Secondly, the deformed mesh after disparity estimation and the previous right image are used to estimate the motion backwardly. Thirdly, the triangle is regarded as the predictive unit, and the prediction which makes the predictive error of a triangle of mesh smallest is chose as the final prediction. So the quality of predicted image can be improved. By the way, a joint motion/disparity estimation based on series mesh deformation, a simple implementation of the proposed method, is proposed. Experimental results show that the proposed method and its simplification, both being integrated into the stereo video coding system, can maintain excellent objective and subjective quality for the decoded image. Under almost the same compression ratio, the average PSNR of decoded image of these two methods are higher about 0.6dB and 0.5dB than that of mesh-based compatible method, respectively. Moreover, the proposed method can provide better subjective quality than MPEG-TPS for decoded image.