Motion Estimation and Structurization for Very Low Bitrate Coding

在本研究中，我们提出了一种新的基于像素匹配的运动估计框架。建议的系统由以下两个中柱组成。第一步是确定变形平面，该变形平面是单个像素的移位帧差的函数，然后为每个像素存储它。第二步是基于一种标准来估计运动，该标准使用在前一步中获得的变形平面上工作的操作员。然后估计给出该标准的最小值(或最大值)的运动矢量。新的框架可以灵活地选择不同的运动模型和算子。本文首先研究了中值算子、非线性加权求和算子、多数判决算子和方向距离变换算子，以有效地处理运动场中的不连续性。其次，我们讨论了仿射运动模型的Hough变换算子。最后，我们提出了一种利用像素匹配信息同时进行运动估计和分割的方法，并研究了该方法在低位视频编码和运动补偿帧内插中的应用。

In this research, we propose a novel framework for motion estimation based on pixel-wise matching. The proposed system consists of following two stapes. First step is the determination of a distortion plane, which is a function of the displaced-frame differemce of a single pixel, and then storing it for each pixel. Second step is the estimation of motion based on a criterion which emplys an operator working on the obtained distortion planes obtained in the previous step. The motion vector that gives the minimum (or mzximum) value for the criterion is then estimated. The new framework is flexible in allowing selection of various motion models and operators.In this thesis, firstly, we study on median operators, nonlinear weighted-summation operators, majority-decision operators, and directional-distance-transformation operators in order to deal effectively with discontinuities in the motion field. Secondly, we discuss a Hough-transform operator for an affine motionn model. Finally, we propose a method which performs motion estimation and segmentation simultaneously using information of pixel-wise matching, and then also study on the application of this method to low-bit-video coding and motion-compensated frame interpolation.