基于样本间最小欧氏距离的多特征融合识别算法研究

摘要

In the process of the target tracking,to improve the accuracy and real-time performance of target image recognition, this paper proposes a multi-feature fusion recognition method based on DS evidence theory and minimum Euclidean distance be?tween samples(E-DS).By image preprocessing and Sobel edge detection to target image,two types of visual features such as the tar?get color and geometry are extracted and normalized to form the target image feature vector;according to DS fusion theory,the mini?mum Euclidean distances between single samples are calculated and the results are used as evidences to construct the basic proba?bility assignment function,combined with DS combination rule,the final recognition results are given.The multi-feature E-DS fu?sion recognition method is applied to the target recognition test,the calculation results show that the average correct recognition rate of E-DS method reaches 95.49%,the highest recognition rate is 97.16%,and the variance of recognition rate between groups is mini?mum,which verifies the applicability of E-DS method.%在目标跟踪过程中,为了提高目标图像识别的准确率和实时性,提出了一种基于样本间最小欧氏距离和DS证据理论的多特征融合识别方法(E-DS).在对目标图像进行预处理和Sobel边缘检测的基础上,提取目标的颜色和几何两类视觉特征,加以归一化,形成目标图像特征向量;根据多特征DS融合理论,以2类单特征样本间的最小欧氏距离作为独立证据构造基本概率分配函数,结合DS证据组合规则,给出最终的识别结果.将多特征E-DS融合识别方法应用于目标识别试验中,计算结果表明此识别方法平均正确识别率达到95.49%,最高正确识别率达到97.16%,且此组间识别率的方差最小,验证了这一方法的有效性.