当前行人再识别的度量算法在计算相似性时主要依据两幅图像自身的判别信息(直接度量),较少依据与两幅图像相关的其它图像的判别信息(间接度量).针对此种情况,文中提出加权融合直接度量和间接度量的度量方法.首先提取图像的局部最大概率特征和突出性颜色名称特征,融合两者作为图像的最终特征.然后分别计算两幅图像的直接相似性和间接相似性,利用序列排序方法对数据库样本进行训练,得到权值参数,从而得到两幅图像的最终相似性.在Market-1501数据库和CUHK03数据库上的实验表明,融合后的度量识别能力明显高于单个度量的识别能力.%The metric algorithm for person re-identification to compute similarity of the image pairs is mostly based on the discriminant information of themselves rather than the discriminant information of other images related to them. Therefore, a metric method is proposed to fuse direct metric and indirect metric by weighing them. Firstly,the local maximal occurrence feature and salient color name feature of the images are extracted,and two features are fused as the final feature of the image. Then, the direct similarity and the indirect similarity of two images are calculated respectively. Finally, the sequence sorting method is further proposed to obtain the weights by training the database samples, and thus the final similarity of two images is acquired. The experimental results on Market-1501 database and CUHK03 database show that the recognition ability of fusion metric is obviously higher than that of the single metric.
展开▼
