Bi-level Relative Information Analysis for Multiple-Shot Person Re-Identification

Abstract

SUMMARYMultiple-shot person re-identification, which is valuable for application in visual surveillance, tackles the problem of building the correspondence between images of the same person from different cameras. It is challenging because of the large within-class variations due to the changeable body appearance and environment and the small between-class differences arising from the possibly similar body shape and clothes style. A novel method named "Bi-level Relative Information Analysis" is proposed in this paper for the issue by treating it as a set-based ranking problem. It creatively designs a relative dissimilarity using set-level neighborhood information, called "Set-level Common-Near-Neighbor Modeling", complementary to the sample-level relative feature "Third-Party Collaborative Representation" which has recently been proven to be quite effective for multiple-shot person re-identification. Experiments implemented on several public benchmark datasets show significant improvements over state-of-the-art methods. key words: Bi-level relative information analysis; multiple-shot person re-identification; visual surveillance

Locality based discriminative measure for multiple-shot human re-identification

“…For final set-based matching, relative position of the sets in the local metric fields is also important. This can be studied by the neighborhood structure information for all sets, which is formulated by "Set-level Common Near Neighbor Modeling (SCNNM)" [16,12,26]. SCNNM quantifies the local situation of the paired sets in each other's neighborhood structures.…”

“…(11), A and B denote two arbitrary sets: Λ is the trade-off parameter between D Symmetric ðA; BÞ and D Asymmetric ðA; BÞ; the "Fixednumber" for the neighborhood size, denoted by N, is suggested to be half of the average number of sets per class [12,16]. Considering symmetry, D Symmetric ðA; BÞ is given by…”

“…Although a sound representation can characterize the human appearance, the complexity of the real-world situation and the subjectivity of the feature hand-designing process inevitably impede the performance enhancement. Other solutions rely on robust measure for the representation to address these problems [9][10][11][12][13][14]. Their inspiring results have been achieved because of improved intra-class compactness and interclass separation.…”

People Re-Identification with Local Distance Comparison Using Learned Metric

“…Recently, some researches such as the work of Li et al [16] focus on transfer learning, in order to deal with overfitting problem caused by small training data. In Li et al's work [17], they sample several images as the "third party" images from another dataset, similar with but different from the probe and gallery sets. The images in the probe and gallery sets are expressed as a collaborative representation based on the "third party" images using a sparse coding method.…”

Person Re-Identification by Common-Near-Neighbor Analysis

“…Synthetic data can help us further elaborate our proposal [19]. We randomly generate three separate, Gaussian-distributed datasets for use as class samples (for a class size of 40).…”