非同质运动图像关键帧耦合特征识别仿真

Keyframe Coupling Feature Recognition Simulation for Non-homogeneous Motion Images

采用当前方法识别非同质运动图像关键帧的耦合特征时,识别所用的时间较长,得到的识别结果与实际不符,存在识别效率低和识别结果准确率低的问题.提出非同质运动图像关键帧耦合特征识别方法,根据噪声属性和像素灰度值属性划分非同质运动图像,结合粗糙集理论融合增强划分子块并最终得到子图.采用不预设K-均值聚簇算法对增强后的非同质运动图像做聚簇处理,通过相似距离提取每个簇中与聚簇中心距离最近的帧作为关键帧,提取关键帧中的时间序列,根据时间序列得到非同质运动图像关键帧的耦合特征,将关键帧耦合特征输入K最近邻混合分类器中,实现非同质运动图像关键帧耦合特征的识别.仿真结果表明,所提方法的识别效率高、识别准确率高.

In current methods, the recognition time is long and the recognition results are inconsistent with the actual result, resulting in low recognition efficiency and low accuracy of recognition result. Therefore, this paper puts forward a method to recognize key frame coupling feature in non-homogeneous motion image. According to the noise attribute and the gray value attribute of pixel, the inhomogeneous motion image was divided, so that sub-images were finally obtained based on rough set theory. The non-predetermined K-means clustering algorithm was used to cluster the enhanced non-homogeneous motion images. Through the similarity distance, the frame which was closest to the cluster center was extracted as the key frame, and the time series in key frame was extracted. On this basis, the coupling characteristic of key frame in the non-homogeneous motion image was obtained. Finally, the coupling feature of key frame was input into K-nearest neighbor hybrid classifier. Thus, the recognition of key frame coupling feature in inhomogeneous motion image was achieved. Simulation results show that the proposed method has high recognition efficiency and high recognition accuracy.