Person re-identification by Local Maximal Occurrence representation and metric learning

By Shengcai Liao, Yang Hu, Xiangyu Zhu, S. Li. Computer Vision and Pattern Recognition, 2014

2,139 citations recorded by the source corpus.

Person re-identification is an important technique towards automatic search of a person's presence in a surveillance video. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this paper, we propose an effective feature representation called Local Maximal Occurrence (LOMO), and a subspace and metric learning method called Cross-view Quadratic Discriminant Analysis (XQDA). The LOMO feature analyzes the horizontal occurrence of local features, and maximizes the occurrence to make a stable representation against viewpoint changes. Besides, to handle illumination variations, we apply the Retinex transform and a scale invariant texture operator. To learn a discriminant metric, we propose to learn a discriminant low dimensional subspace by cross-view quadratic discriminant analysis, and simultaneously, a QDA metric is learned on the derived subspace. We also present a practical computation method for XQDA, as well as its regularization. Experiments on four challenging person re-identification databases, VIPeR, QMUL GRID, CUHK Campus, and CUHK03, show that the proposed method improves the state-of-the-art rank-1 identification rates by 2.2%, 4.88%, 28.91%, and 31.55% on the four databases, respectively.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/f0822cc18e502e15a8c153773492934201e4c8f9

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Transfer Learning for Person Re-Identification (2016, background): “Earlier works focus on either designing viewinsensitive feature representations [2], [3], or learning an effective distance metric [4], [5], or both [6].”
Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification (2016, methodology): “Liao et al . improved the KISSME method by learning a discriminant low dimensional subspace [21] based on the LOMO features.”
Multi-Task Learning with Low Rank Attribute Embedding for Multi-Camera Person Re-Identification (2018, methodology): “We compare MTL-LORAE with some recent supervised learning-based methods, including KISSME [11], kLFDA [12], KCCA [72], LOMO+XQDA [73], TSR [74], EPKFM [75]and MLAPG [76], as shown in Table 4.”
Gated Siamese Convolutional Neural Network Architecture for Human Re-identification (2016, methodology): “This strategy is inspired by the technique introduced in [24] in which the features along a single row is pooled to make the final feature map to a shape (number of rows) ×1.”
Tracking the Untrackable: Learning to Track Multiple Cues with Long-Term Dependencies (2017, neutral): “3 95 LX [43] 46.”
RGB-Infrared Cross-Modality Person Re-Identification via Joint Pixel and Feature Alignment (2019, methodology): “Existing methods can be grouped into hand-crafted descriptors [22, 36, 19, 34], metric learning methods [43, 16, 20] and deep learning algorithms [42, 11, 28, 21, 44, 30, 15, 35, 33, 18, 17, 1, 31].”
VERI-Wild: A Large Dataset and a New Method for Vehicle Re-Identification in the Wild (2019, neutral): “Methods mAP R=1 R=5 mAP R=1 R=5 LOMO [8] - 18.”
Multi-Task Learning with Low Rank Attribute Embedding for Person Re-Identification (2015, methodology): “We compare MTL-LORAE with 4 best-performing methods, including 2 recent ones: LOMO+XQDA (LX) [32] and TSR [40], as shown in Table 3.”
Scalable Person Re-identification on Supervised Smoothed Manifold (2017, methodology): “Current research interests can be coarsely divided into two mainstreams: 1) those focus on designing robust visual descriptors [55, 13, 37, 26, 64] to accurately model the appearance of person; 2) those seek for a discriminative metric [65, 25, 54, 28, 18], under which instances of the same identity”
Bi-Directional Center-Constrained Top-Ranking for Visible Thermal Person Re-Identification (2020, methodology): “Current Re-ID methods can be roughly categorized into two aspects: feature representation learning [23]–[25] and distance metric learning [26]–[29].”
Adaptive Transfer Network for Cross-Domain Person Re-Identification (2019, background): “Person re-identification is the task of matching a probe pedestrian image from a large-scale gallery collected by non-overlapping camera networks at diverse locations [17, 35, 18].”
Mask-Guided Contrastive Attention Model for Person Re-identification (2018, methodology): “Then we compute the distance between probe and gallery samples with several classic metrics, including conventional Eulidean distance, XQDA [26], KISSME [21] and the recently proposed Re-ranking methods [52].”
Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-identification (2018, background): “thousands for typical low-level feature representations) [56, 29, 10, 57]; (2) Consistently predicting individually all the attributes is a difficult task when the labelled training data is sparse and person images have low quality as mostly in person re-id datasets, that is, interattribute discrimi”
Dual Attention Matching Network for Context-Aware Feature Sequence Based Person Re-identification (2018, background): “…usually describes each pedestrian image or video footage with a single feature vector ﬁrstly and then match them in a task-speciﬁc metric space, where the feature vectors of same pedestrian are expected to have smaller dis-( tances than that of different pedestrians, e.g ., [20, 34, 21, 64, 54].”
Multi-scale Deep Learning Architectures for Person Re-identification (2017, methodology): “Typically, person re-id is formulated only as a veriﬁcation task [24, 32, 57, 59].”
Efficient PSD Constrained Asymmetric Metric Learning for Person Re-Identification (2015, methodology): “15% rank-1 identification rates with the labeled bounding boxes and the automatically detected bounding boxes, respectively, which clearly outperform the state-of-the-art method XQDA [14], with an improvement of 5.”