现场勘验刀具图像特征描述与识别

Characteristics description and recognition of knife images in crime scene investigation

针对人工录入现勘刀具图像耗时耗力、准确率低的问题,提出一种基于目标形状特征的现勘刀具图像识别算法。在分析现勘刀具图像特色数据基础上,总结现勘刀具图像的典型内容特性;利用基于结构森林的边缘检测和聚类的方法对目标定位和轮廓提取,再结合包含目标轮廓的最小外接矩形给出刀尖角度的计算方法,并与长宽比、矩形度、圆形度构建识别刀目标的形状特征向量,将该特征向量输入支持向量机建立二分类模型,实现对现勘刀具图像的识别。实验结果表明,该算法性能明显优于其他特征提取算法,且对现勘刀具图像识别准确率达到94.61%,可有效表征刀具图像的内容。

Recording knife images during crime scene investigation is time-consuming,labour-intensive,and sometimes inaccurate due to human error.An effective method for crime scene investigation knife images recognition is proposed in this paper using the shape characteristics.Based on the analysis of the characteristic data of the crime scene investigation knife images,the typical content characteristics of the crime scene investigation image are summarized;the edge detection based on the structure forest and clustering algorithm are used to locate and extract the target,and then combined with the smallest circumscribed rectangle containing the knife contour,the calculation method of the angle of the knife tip is established.The shape feature vector identifying the knife target is constructed with the aspect ratio,rectangularity,circularity and the angle of the knife tip.The feature vector is input to the support vector machine to establish a binary classification model to realize the recognition of the image of crime scene investigation knife image.The method proposed in this paper can achieve the highest recognition accuracy,reaching 94.61%on crime scene investigation knife images.