Discriminant quaternion local binary pattern embedding for person re-identification through prototype formation and color categorization
Engineering Applications of Artificial Intelligence
Re-identifying objects is one of the fundamental elements for visual surveillance, in the sense that images of the same object at different time or places should be assigned with the same label. In this work, we propose a new embedding scheme for person re-identification under nonoverlapping target cameras. Inspired by the prototype approach derived from cognition field, we propose to use prototype images as a reference set to achieve a discriminative representation of a person's appearance. To enhance the discrimination between different persons, we learn a linear subspace in a training phase during which person correspondences are assumed to be known. The robustness of the algorithm against results that are counterintuitive to a human operator is improved by proposing the Color Categorization procedure. By doing so, our method becomes very flexible when tracing a person in a camera network even under large illumination changes. The proposed framework was tested on VIPeR, the most challenging dataset for person re-identification. Results confirm that our method outperforms the state of the art techniques.