Person re-identification combining deep features and attribute detection

Attributes-Based Re-Identification is a way of identifying individuals when presented with multiple pictures taken under varying conditions. The method typically builds a classifier to detect the presence of certain appearance characteristics in an image, and creates feature descriptors based on the output of the classifier. We improve attribute detection through spatial segregation of a person’s limbs using a skeleton prediction method. After a skeleton has been predicted, it is used to crop the image into three parts - top, middle and bottom. We then pass these images to an attribute prediction network to generate robust feature descriptors. We evaluate the performance of our method on the VIPeR, PRID2011 and i-LIDS data sets, comparing our results against the state-of-the-art to demonstrate competitive overall matching performance.