基于图像显著特征的非重叠视域行人再识别

Pedestrian re-identification based on salient features in non-overlapping areas

显著特征是视觉识别中重要的鉴别信息,文中提出了两种显著特征来表征行人图像,一种基于快速鲁棒显著特征,另一种是基于主成分尺度不变显著特征。分别在图像的感兴趣区域中提取SURF(Speeded Up Robust Features)特征和PCA-SIFT(Principal Component Analysis-Scale Invariant Feature Transform)特征描述行人图像的纹理信息结合色彩直方图信息构成特征空间。通过待测图像与其余图像相应区域相似性估计来提取每幅图像中的特征显著性分布,得到图像的显著特征。最后在GRID(Underground Re-Identification Dataset)库上大量的实验验证了两种显著特征表示方法,并在行人再匹配中的获得了较SIFT(Scale Invariant Feature Transform)特征更高的识别效率和识别精度。

Salient feature in images always provides valuable information for pattern recognition, therefore two kinds of salient features to represent pedestrians are presented. A speeded up robust feature （SURF） and a principal component analysis-scale invariant feature transform （PCA-SIFT） feature are used to de- scribe texture of person image, respectively, then texture information and color histogram are used to con- struct the feature space. In the feature space, salient features are detected according to the difference be- tween person images. The effectiveness of salient features in people re-identification is validated on the widely used GRID dataset. Experimental results show that it outperforms other approaches in the accuracy and the efficiency using SURF and PCA-SIFT based on salient features.