基于光照不变性颜色特征的行人再识别方法

The method of person re-identification based on illumination invariant color feature

针对光照变化对行人的颜色外观有很大影响导致行人再识别率低的问题,提出了一种基于光照不变性颜色特征的行人再识别方法。使用颜色属性转换(Color Characteristic Transfer,CCT)把不同光照条件下的行人图像统一转换到相近亮度,把转换后的图像从像素值变化大的RGB空间转换到可以使像素值变化相对较小的颜色空间中去,提取出颜色直方图作为光照不变性特征。采用跨视图二次鉴别分析(Cross-view Quadratic Discriminant Analysis,XQDA)算法进行行人图像特征向量间的相似性度量。通过在公共数据集VIPeR和CUHK01上的实验,表明了文中所提出的特征对光照变化有一定的鲁棒性,且算法与相关的行人再识别方法相比,识别精度得到了进一步提升,具有更高的实际应用价值。

To solve the problem that the change of illumination has a great influence on the color appearance of pedestrians,which could lead to the low recognition rate,the illumination invariant color feature for person re-identification is proposed.Color characteristic transfer is used to transform the pedestrian images under different illumination conditions to the similar brightness.The images are converted from RGB space with large pixel value changes to the color spaces with pixel relatively smaller changes.The color histograms are extracted as the illumination invariant color feature.The Cross-view Quadratic Discriminant Analysis(XQDA)algorithm is used here to measure the similarity between feature vectors of pedestrian images.Experiments on public datasets VIPeR and CUHK01 demonstrate that the feature proposed is robust to illumination changes,and the recognition accuracy of the algorithm is further improved compared with the related person re-identification methods,which has higher practical application value.