A rank-order distance based clustering algorithm for face tagging

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

Section: Merits Of Cnnmmentioning

confidence: 98%

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

Wei

Minoh

2013

“…As distance metric, we use Rank-order distance [26] which has been demonstrated as a better density measurement than commonly used Euclidean distance [17]. We train a temporal ConvNet using UCF101 dataset [11] (split 1) following the approach proposed in [13] except that we insert a normalization layer between pool2 layer and conv3 layer.…”

Section: Methodsmentioning

confidence: 99%

Automatic Retrieval of Action Video Shots from the Web Using Density-Based Cluster Analysis and Outlier Detection

Yanai

2016

SUMMARYIn this paper, we introduce a fully automatic approach to construct action datasets from noisy Web video search results. The idea is based on combining cluster structure analysis and density-based outlier detection. For a specific action concept, first, we download its Web top search videos and segment them into video shots. We then organize these shots into subsets using density-based hierarchy clustering. For each set, we rank its shots by their outlier degrees which are determined as their isolatedness with respect to their surroundings. Finally, we collect high ranked shots as training data for the action concept. We demonstrate that with action models trained by our data, we can obtain promising precision rates in the task of action classification while offering the advantage of fully automatic, scalable learning. Experiment results on UCF11, a challenging action dataset, show the effectiveness of our method.

“…There exists one similar method, Zhu et al's Rank-Order distance, originally designed for clustering during face tagging [17]. Although this method is capable of solving the samples' non-uniform distribution problem during clustering by rank order quantization, yet it may fail to re-identify the human image data with large intra-class variations and small inter-class differences, as the general case in the real world.…”

Section: Comparison Between Cnnm and Its Analogue 221 Weakness Of Rmentioning

confidence: 99%

“…To overcome the weakness of this method in discriminating real-world human image data, we initially propose to quantify the situation of the local neighborhood structure of one sample in the neighborhood structure of the other sample into a dissimilarity in the learned metric space. To show the advantage of our proposal, Rank-Order distance [17] provides a good contrast. Rank-Order distance uses rank order quantization to solve the samples' non-uniform distribution problem during clustering.…”

Section: Background and Related Workmentioning

confidence: 99%

“…Zhu et al used experiments to demonstrate the robustness of Rank-Order distance during clustering. For clusters with varying density/shape/size, they plotted all the edges between points whose Rank-Order distances are smaller than a given threshold [17]. We use experiments to visualize the advantage of CNNM evaluated by recognition/rankings.…”

Section: Advantage Of Cnnmmentioning

confidence: 99%

See 1 more Smart Citation

Person Re-Identification by Common-Near-Neighbor Analysis

Wei

Yinghui²,

et al. 2014

SUMMARYRe-identifying the same person in different images is a distinct challenge for visual surveillance systems. Building an accurate correspondence between highly variable images requires a suitable dissimilarity measure. To date, most existing measures have used adapted distance based on a learned metric. Unfortunately, real-world human image data, which tends to show large intra-class variations and small inter-class differences, continues to prevent these measures from achieving satisfactory re-identification performance. Recognizing neighboring distribution can provide additional useful information to help tackle the deviation of the to-be-measured samples, we propose a novel dissimilarity measure from the neighborhood-wise relative information perspective, which can deliver the effectiveness of those well-distributed samples to the badly-distributed samples to make intra-class dissimilarities smaller than inter-class dissimilarities, in a learned discriminative space. The effectiveness of this method is demonstrated by explanation and experimentation.