Person re‐identification by modelling principal component analysis coefficients of image dissimilarities
Niki Martinel, Christian Micheloni · Electronics Letters · 2014
Signature‐based matching has been the dominant choice for state‐of‐the‐art person re‐identification across multiple disjoint cameras. An approach that exploits image dissimilarities is proposed, treating re‐identification as a binary classification problem. To achieve the objective, the person re‐identification problem is addressed as follows: (i) first, compute the image dissimilarity between a pair of images acquired from two disjoint cameras; (ii) then learn the linear subspace where the image dissimilarities lie in an unsupervised fashion and (iii) lastly train a binary classifier in the linear subspace to discriminate between image dissimilarities computed for a positive pair (images are for the same person) and a negative pair (images are for different persons). An approach on two publicly available benchmark datasets is evaluated and compared with state‐of‐the‐art methods for person re‐identification.