Locally Aligned Feature Transforms across Views

Abstract

In this paper, we propose a new approach for matching images observed in different camera views with complex cross-view transforms and apply it to person reidentification. It jointly partitions the image spaces of two camera views into different configurations according to the similarity of cross-view transforms. The visual features of an image pair from different views are first locally aligned by being projected to a common feature space and then matched with softly assigned metrics which are locally optimized. The features optimal for recognizing identities are different from those for clustering cross-view transforms. They are jointly learned by utilizing sparsityinducing norm and information theoretical regularization. This approach can be generalized to the settings where test images are from new camera views, not the same as those in the training set. Extensive experiments are conducted on public datasets and our own dataset. Comparisons with the state-of-the-art metric learning and person re-identification methods show the superior performance of our approach.

Multiple-Shot People Re-Identification by Patch-Wise Learning

“…An experimental setting similar to that used in previous studies [5], [24] is employed here. For each dataset, images of half of all the people are used as the training set; images of the remaining people are used as the testing set.…”

“…These images were selected by maximizing the variance with respect to the resolution, illumination, occlusion, and posture [16], which makes the dataset a very challenging one. Similar as in [5], [7], [9], we choose all the 72 people in the experiments. The 20 people, captured by the same camera view, are also included with the consideration of challenges on wide range of posture, real surveillance footage, and wide range of resolution.…”

“…These methods may easily fail to suppress the variance issue simultaneously in the case where both of the causes are severe. Li et al projected the feature representation taken from different views to align them locally on different feature subspaces [5]. Each subspace was associated with different specific properties of feature representation.…”

“…The metric learning method is effective, however it generally fails to deal with both issues together. One efficient method is to weaken the effects of the non-rigid deformation issue, such as projecting feature representations to common subspaces or formulating them in multi-modal distribution [5], [9], and focus on dealing with the varying illumination issue. In the second group, the approaches are efficient in comparing the corresponding parts against the non-rigid deformation, whereas they rely on hand-crafting the effective features and extracting parts information.…”

A Study on Person Re-Identification System using Enhanced RNN

“…[24] show that the color as a single cue under extremely variable imaging conditions has relatively good performance in identifying persons, and present a novel illumination invariant feature representation which is based on the log chromaticity (log) color space. The existing features include: local binary patterns [25,11,12,20,14] , salient color names [26] , variations on color histograms [25,11,12,20,14,29] , Gabor features [14] , and local patches [28] .…”