BiCov: a novel image representation for person re-identification and face verification

By Bingpeng Ma, Yu Su, Frédéric Jurie. British Machine Vision Conference, 2012

297 citations recorded by the source corpus.

This paper proposes a novel image representation which can properly handle both background and illumination variations. It is therefore adapted to the person/face reidentification tasks, avoiding the use of any additional pre-processing steps such as foreground-background separation or face and body part segmentation. This novel representation relies on the combination of Biologically Inspired Features (BIF) and covariance descriptors used to compute the similarity of the BIF features at neighboring scales. Hence, we will refer to it as the BiCov representation. To show the effectiveness of BiCov, this paper conducts experiments on two person re-identification tasks (VIPeR and ETHZ) and one face verification task (LFW), on which it improves the current state-of-the-art performance.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/2aacfcdc5d06c86901852f7b666d17e97822ba23

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

1. Yi Yang
2. Liang Zheng
3. Xiaogang Wang
4. Qi Tian
5. Luc Van Gool
6. Dacheng Tao
7. Ming-Hsuan Yang
8. Tao A. Xiang

In Signal Processing

1. Lei Zhang
2. Yi Yang
3. Wangmeng Zuo
4. Liang Zheng
5. Jiwen Lu
6. David Dapeng Zhang
7. Wanli Ouyang
8. Zhangyang Wang

Citations in the corpus

• PersonNet: Person Re-identification with Deep Convolutional Neural Networks (2016, methodology): “c) Competitors: We compare our model with the following state-of-the-art approaches: SDALF [11], ELF [29], LMNN [17], ITML [34], LDM [35], eSDC [7], SalMatch [10], Generic Metric [41], Mid-Level Filter (MLF) [4], eBiCov [43], PCCA [16], LADF [9], kLFDA [6], rPCCA [6], RDC [13], RankSVM [44], Metric ”
• Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification (2016, methodology): “We report the CMC curves and rank−1 accuracies of local Fisher discriminant analysis (LF) [29], pairwise constrained component analysis (PC− CA) [28], aPRDC [23], PRDC [43], enriched BiCov (eBi− Cov) [25], PRSVM [2], and ELF [10], saliency matching (SalMatch) [40], patch matching (PatMatch) [40], lo”
• ABD-Net: Attentive but Diverse Person Re-Identification (2019, background): “We mainly review the former where both handcrafted features [18, 19, 20, 21] and learned features [22, 23, 4, 24, 25] were studied.”
• Person Re-identification: Past, Present and Future (2016, methodology): “This multi-shot matching strategy is adopted by later works [97], [98].”
• Similarity Learning with Spatial Constraints for Person Re-identification (2016, background): “The former type adopts the offthe-shelf distance metric such as Euclidean distance [10], Bhattacharyya distance[6], and covariance distance [22, 1], while the latter type tries to exploit the inherent invariance between image pairs [12, 27, 13, 24].”
• Person Re-identification by Salience Matching (2013, methodology): “(a) On VIPeR dataset, our approach (PatMatch and SalMatch) is compared with benchmarking methods including SDALF [5], CPS [3], eBiCov [16], eSDC [22], ELF [8], PRSVM [19], PRDC [23], LMNN-R [4], aPRDC [13], and PCCA [18]; (b) On CUHK Campus dataset, our approach is compared with L1-norm distance, L2”
• A Siamese Long Short-Term Memory Architecture for Human Re-identification (2016, background): “Most of the works on human re-identification concentrates on either developing a new feature representation [9, 27, 33, 36, 58, 59, 63] or learning a new distance”
• Unsupervised Salience Learning for Person Re-identification (2013, methodology): “As shown in Figure 7, our approach outperforms the three benchmarking methods, PLS, SDALF and eBiCov[21] on all three sequences.”
• Brownian descriptor: A rich meta-feature for appearance matching (2014, methodology): “We evaluate our approach on tracking related applications, demonstrating that the Brownian descriptor outperforms the classical covariance in terms of matching accuracy and efficiency.”
• A survey of approaches and trends in person re-identification (2014, unknown): “Covariance matrices are also adopted in [36] but the underlying features are Gabor filter responsemagnitude images extracted from different spatial scales (BiCov).”
• Learning Mid-level Filters for Person Re-identification (2014, methodology): “We also compare our approach with benchmarking methods including SDALF [5], eBiCov [17], eSDC [28], PRDC [29], aPRDC [15], PCCA [18], KISSME [9], LF [19], SalMatch[27] and LADF[14].”
• Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in Vitro (2017, background): “Some pioneering works focus on finding discriminative handcrafted features [22, 23, 20].”
• An improved deep learning architecture for person re-identification (2015, methodology): “1, we compare our method against local Fisher discriminant analysis (LF) [23], PCCA [22], aPRDC [18], PRDC [32], eBiCov [19], LMNNR [1], PRSVM [2], and ELF [7].”
• Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function (2016, methodology): “Features that have been used for the person re-id task include color histograms and their variants [41, 20, 21, 28, 23, 46], local binary patterns(LBP) [41, 20, 21, 28, 23, 46], Gabor features [23], color names [44], and other visual appearance or contextual cues [3].”