提高视频对象分割的实时性及多对象分割

视频对象的分割目前存在两个难题,一是如何提高分割的实时性,二是对存在互遮挡的多个对象如何进行有效分割。对前者分割提出了基于自适应跳帧的分割算法和基于CNN(细胞神经网络)的实时分割法,对后者提出了基于时空曲线演化的多视频对象分割和Bayes最小误差判断的方法。仿真实验表明,这些方法是有效的,对这两个难题的最终解决和实际应用指出了新的途径。

Real-time Realization of Video Object Segmentation and Multiple Object Extraction

Video object segmentation exists two difficult problems:how to segment object in real-time and extract multiple objects while the occlusion is emerged. To the first problem, a segmentation algorithm based on adaptively skipped frames and CNN (Cellular Neural Network) are presented, and to the second, it is proposed a novel multiple object segmentation algorithm based on spatial-temporal curve evolution and Bayes classification for minimum error. The experiment results show that the algorithms are effective.