Advances In Face Recognition Research Articles

Face Recognition from Small Datasets using Kernel Selection of Gabor Features

Recent advances in face recognition are mostly based on deep methods that require large datasets for training. This paper presents a novel method that combines Gabor-features, feature and kernel to achieve comparable performance on smaller datasets.The paper compares different feature methods in this context.The problem tackled in this paper is achieving accurate face recognition with limited computational resources. By “limited computational resources we mean low computational power (i.e. memory, CPU ops) during both system training and evaluation. Noted that we are not competing against deep learning systems in term of accuracy but we provided a middle ground between hand-coded fast feature extraction and learning based deep learning in terms of both speed and accuracy.To achieve this goal, we propose “kernel selection as the main method to reduce the dimensionality of the classification problem faced by the final classifier in the FR system. Kernel is the process of eliminating less important Gabor kernels for classification while keeping the level of accuracy achievable. Kernel differs from traditional feature in measuring the value of complete kernels consisting of several features together. Because of its structured nature, Kernel has the advantage of eliminating the need to evaluate complete Gabor kernels reducing the computational cost of the system compared with traditional feature methods.

Face recognition with dense supervision

Recent advances in face recognition mostly concentrate on designing more discriminative loss functions or adding normalization on features/weights to make a single feature more accurate. In this work, inspired by the frequently used multi-patch ensemble method for face recognition and part-based models for person re-identification, we propose a novel training strategy to enhance the discriminability of deeply learned feature from another perspective, namely learning with dense supervision. The main idea is to apply multiple classification losses on top of multiple component features extracted from a single network. Ideally, each component feature is expected to be accurate and have low correlation with the others. To this end, we first design a metric called feature consistency to evaluate the correlation between one component feature and the others, which is defined as the sum of distances between one component feature and the others, where the distance here is measured with KL divergence between corresponding softmax probabilities. Then we use feature consistency to select which component features to sample for one learning pass by importance sampling. The dense supervision significantly outperforms the single supervision baseline and even performs on par with its multi-patch ensemble counterpart which has much more parameters (9×). Our experimental results match state-of-the-art performance on LFW, YTF, MegaFace and surpass the others on LFW BLUFR and VGGFace2 pose protocol, thereby achieving state-of-the-art. Specially, results on VGGFace2 also show the superiority of the dense supervision on cross-pose face matching.

Spectral-Spatial Feature Extraction and Classification by ANN Supervised With Center Loss in Hyperspectral Imagery

In this paper, we propose a spectral-spatial feature extraction and classification framework based on artificial neuron network (ANN) in the context of hyperspectral imagery. With limited labeled samples, only spectral information is exploited for training and spatial context is integrated posteriorly at the testing stage. Taking advantage of recent advances in face recognition, a joint supervision symbol that combines softmax loss and center loss is adopted to train the proposed network, by which intra-class features are gathered while inter-class variations are enlarged. Based on the learned architecture, the extracted spectrum-based features are classified by a center classifier. Moreover, to fuse the spectral and spatial information, an adaptive spectral-spatial center classifier is developed, where multiscale neighborhoods are considered simultaneously, and the final label is determined using an adaptive voting strategy. Finally, experimental results on three well-known datasets validate the effectiveness of the proposed methods compared with the state-of-the-art approaches.

A weighted probabilistic approach to face recognition from multiple images and video sequences

To date, advances in face recognition have been dominated by the design of algorithms that do recognition from a single test image. Recently, an obvious but important question has been put forward. Will the recognition results of such approaches be generally improved when using multiple images or video sequences? To test this, we extend the formulation of a probabilistic appearance-based face recognition approach (which was originally defined to do recognition from a single still) to work with multiple images and video sequences. In our algorithm, as it is the case in most appearance-based approaches, we will need to use a feature extraction algorithm to find those features that best describe and discriminate among face images of distinct people. We will show that regardless of the algorithm used, the recognition results improve considerably when one uses a video sequence rather than a single still. Hence, a positive answer to our question (in the general sense) seems reasonable. The probabilistic algorithm we propose in this paper is robust to partial occlusions, orientation and expression changes, and does not require of a precise localization of the face or facial features. We will also show how these problems are more easily solved when one uses a video sequence rather than a single image for testing. The limitations of our algorithm will also be discussed. Understanding the limitations of current techniques when applied to video is important, because it helps identify those weak points that require further consideration.