刘颖 谢晓飞 来毅 李芸 柴豪骏

摘  要： 刑侦现勘图像存在质量差、模糊、目标复杂且分辨力低等缺点，这些缺点导致图像匹配时出现效果差、耗时长等问题，针对该问题提出一种基于网格运动统计（GMS）的特征匹配算法。该算法首先使用网格对图像均匀分割，然后在网格邻域中统计额外支持特征的数量，其次分别计算正确与错误匹配的概率以构造评分函数，最后通过评分函数和评分阈值来区分正确与错误匹配。实验结果表明，该算法对刑侦现勘图像能够实现快速准确的匹配，且该算法对模糊人脸和高纹理的刑侦现勘图像都有较好的匹配效果。

关键词： 刑侦现勘; 网格运动统计; 特征匹配; 平滑运动; 图像分割; 比较验证

中图分类号： TN911.73?34                         文献标识码： A                       文章编号： 1004?373X（2020）06?0178?05

An image matching method for criminal investigation and scene survey

based on grid motion statistics

LIU Ying1，2， XIE Xiaofei2， LAI Yi1，2， LI Yun3， CHAI Haojun2

（1. Key Laboratory of Electronic Information Scene Investigation and Application Technology， Ministry of Public Security， Xian 710121， China;

2. School of Communications and Information Engineering， Xian University of Posts and Telecommunications， Xian 710121， China;

3. CAS Xian Institute of Optics and Precision Mechanics， Xian 710121， China）

Abstract： There are some disadvantages in the image of criminal investigation and scene survey， such as poor quality， fuzzy， complex target and low resolution， resulting in poor effect and long time?consumption in image matching. A feature matching algorithm based on grid motion statistics （GMS） is proposed for above problems. By means of this algorithm， the image is segmented uniformly by grid， and the number of additional supporting features is counted in the grid neighborhood; then the probability of correct and wrong matching is calculated to construct the scoring function， and the correct and wrong matching is distinguished by means of the scoring function and scoring threshold. The experimental results show that the algorithm can achieve fast and accurate matching images of the criminal investigation and scene survey， and the algorithm has a certain matching effect on both fuzzy face and high?texture criminal investigation and scene survey images.

Keywords： criminal investigation and scene survey; grid motion statistics; feature matching; smooth motion; image segmentation; comparison verification

0  引  言

在刑事侦查中，公安人员需要将获取的图像与刑侦现勘数据库中的图像进行对比分析。但由于某些客观的特殊因素的影响，人眼无法分析判断，得出正确的结论。因此，将图像特征匹配技术用于刑侦图像，辅助公安人员分析判断，从而得到正确的信息，提高案件侦查的效率，这具有很重要的现实意义。

近些年以来，许多技术[1?6]使用匹配分布约束区分正确和错误匹配，但这些算法会产生复杂的很难理解的平滑约束，而且最小化的代价很大。文献[2，5]通过一致性约束来获得更快速准确的特征匹配，但基于一致性的特征匹配计算代价大，实现起来很复杂。一些关于特征匹配的经……