论文部分内容阅读
本文提出了基于视差和帧差运动检测的立体视频对象分割算法,该算法结合了视差图分割和多次帧差运动分割的优点,首先对视差图进行分割得到处于不同视差层的目标初级分割模板。然后在模板区域内进行多次帧差运动检测,并且利用边缘检测修正对象边缘,最终得到精确运动目标。在立体视频对象分割的基础上,在MPEG-4的模型上实现了基于对象的立体视频编码。即左通道对象采用普通的MPEG-4编码,右通道对象进行两种方式的预测:一种是基于右通道先前帧图像的MCP(Motion compensation pred iction)方式,另一种是基于左通道图像对应帧的DCP(D isparitycompensation pred iction)方式,然后从中选择预测误差较小的一种。仿真结果表明该方法可以得到良好的重建图像质量,兼容国际标准,易于实现。
This paper proposes a stereo video object segmentation algorithm based on parallax and frame difference motion detection. The algorithm combines the advantages of parallax-image segmentation and multi-frame motion segmentation. Firstly, the parallax map is segmented to get the target primary segmentation template . Then, the frame difference motion detection is performed several times in the template area, and the edge of the object is corrected by the edge detection to finally obtain the precise moving target. Based on the segmentation of stereo video object, object-based stereo video coding is implemented on the MPEG-4 model. That is, the left channel object adopts ordinary MPEG-4 encoding and the right channel object performs prediction in two ways: one is based on the Motion Compensation Prediction (MCP) mode of the previous frame image of the right channel, and the other is based on the left channel image correspondence Frame DCP (D isparitycompensationprediction) way, and then select a smaller prediction error. The simulation results show that this method can get a good reconstructed image quality, compatible with international standards and easy to implement.