视频序列中视觉显著性图像区域自动提取仿真  

Auto-Extraction Simulation of Visual Saliency Image Region in Video Sequence

作　　者：李博[1] LI Bo(China Medical University,Liaoning Shenyang 110000,China)

机构地区：[1]中国医科大学,辽宁沈阳110000

出　　处：《计算机仿真》2021年第7期447-450,共4页Computer Simulation

摘　　要：针对传统显著性图像区域提取方法存在目标提取不准确、结果不稳定等问题,提出一种融合局部与全局特征的视频序列中图像显著性区域提取方法。首先对视频序列图像进行互不重叠的分块,当所有图像块都经过分块并投影到高维空间后,依据单独特征对应显著性区域的规律得到基于全局特征的显著性区域。根据邻域内中心块与其它图像块的灰度差异性,获得基于局部特征的图像显著性区域。最后计算这两个显著图的对比度特征和分布性特征,实现对原始视频序列图像的显著性区域提取。实验结果证明,能够准确提取出视频序列中的图像显著性区域,提取结果更稳定。Aiming at the problems of inaccurate target extraction and unstable results in traditional saliency image region extraction methods, this article proposes a method of extracting the salient area of image in video sequence integrating local feature and global feature. Firstly, the video sequence images were divided into non-overlapping blocks. After all the image blocks were partitioned and projected into the high-dimensional space, the salient area based on global feature was obtained according to the law of salient area corresponding to individual feature. According to the gray difference between the center block and other blocks in the neighborhood, the salient area of image based on local features was obtained. Finally, the contrast characteristics and distribution characteristics of significant graphs were calculated. Thus, the extraction for salient region in original video sequence image was achieved. Experimental results show that the proposed method can accurately extract the salient region of image in video sequence, and the extraction result is more stable.

关 键 词：视频序列 显著性区域 图像块 自适应提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]