论文部分内容阅读
文中提出一种任意形状区域的图像描述方法 ,并根据该方法进行图像序列的运动估计。首先 ,对初始帧图像采用保边界平滑 ,然后 ,对平滑帧进行分水岭分割。根据标记图像 ,将图像以区域邻接图的方式进行存储。为解决过度分割的问题 ,给出了区域的灰度特性和边界强度及区域尺寸相结合的融合准则。最后 ,每个区域采用仿射运动模型及非线性最小均方法进行参数优化。得到的运动矢量场同基于像素点的估计方法相比具有更高的一致性和抗干扰性 ,运动补偿差图像的峰值信噪比得以提高。
In this paper, an image description method of arbitrary shape area is proposed, and the motion estimation of image sequence is carried out according to the method. First of all, the initial frame image is insured with boundary smoothing, and then, the smooth frame is subjected to watershed segmentation. According to the mark image, the image is stored as a region adjoining figure. In order to solve the problem of over-segmentation, the fusion criterion of the combination of the regional gray-scale characteristics and the boundary intensity and the regional size is given. Finally, affine motion models and nonlinear least mean squares method are used for parameter optimization in each region. The obtained motion vector field has higher consistency and anti-interference than the pixel-based estimation method, and the peak signal-to-noise ratio of the motion compensated difference image is improved.