Micro-expressions play an essential part in understanding non-verbal communication and deceit detection. They are involuntary, brief facial movements that are shown when a person is trying to conceal something. Automatic analysis of micro-expression is challenging due to their low amplitude and to their short duration (they occur as fast as 1/15 to 1/25 of a second). We propose a fully micro-expression analysis system consisting of a high-speed image acquisition setup and a software framework which can detect the frames when the micro-expressions occurred as well as determine the type of the emerged expression. The detection and classification methods use fast and simple motion descriptors based on absolute image differences. The recognition module it only involves the computation of several 2D Gaussian probabilities. The software framework was tested on two publicly available high speed micro-expression databases and the whole system was used to acquire new data. The experiments we performed show that our solution outperforms state of the art works which use more complex and computationally intensive descriptors.

The human face holds a privileged position in multi-disciplinary research as it conveys much information—demographical attributes (age, race, gender, ethnicity), social signals, emotion expression, and so forth. Studies have shown that due to the distribution of ethnicity/race in training datasets, biometric algorithms suffer from “cross race effect”—their performance is better on subjects closer to the “country of origin” of the algorithm. The contributions of this paper are two-fold: (a) first, we gathered, annotated and made public a large-scale database of (over 175,000) facial images by automatically crawling the Internet for celebrities’ images belonging to various ethnicity/races, and (b) we trained and compared four state of the art convolutional neural networks on the problem of race and ethnicity classification. To the best of our knowledge, this is the largest, data-balanced, publicly-available face database annotated with race and ethnicity information. We also studied the impact of various face traits and image characteristics on the race/ethnicity deep learning classification methods and compared the obtained results with the ones extracted from psychological studies and anthropomorphic studies. Extensive tests were performed in order to determine the facial features to which the networks are sensitive to. These tests and a recognition rate of 96.64% on the problem of human race classification demonstrate the effectiveness of the proposed solution.

The accurate extraction and measurement of eye features is crucial to a variety of domains, including human-computer interaction, biometry, and medical research. This paper presents a fast and accurate method for extracting multiple features around the eyes: the center of the pupil, the iris radius, and the external shape of the eye. These features are extracted using a multistage algorithm. On the first stage the pupil center is localized using a fast circular symmetry detector and the iris radius is computed using radial gradient projections, and on the second stage the external shape of the eye (of the eyelids) is determined through a Monte Carlo sampling framework based on both color and shape information. Extensive experiments performed on a different dataset demonstrate the effectiveness of our approach. In addition, this work provides eye annotation data for a publicly-available database.

This paper presents a system that automatically extracts the position of the eyeglasses and the accurate shape and size of the frame lenses in facial images. The novelty brought by this paper consists in three key contributions. The first one is an original model for representing the shape of the eyeglasses lens, using Fourier descriptors. The second one is a method for generating the search space starting from a finite, relatively small number of representative lens shapes based on Fourier morphing. Finally, we propose an accurate lens contour extraction algorithm using a multi-stage Monte Carlo sampling technique. Multiple experiments demonstrate the effectiveness of our approach.

Large-scale deep learning models have achieved impressive results on a variety of tasks; however, their deployment on edge or mobile devices is still a challenge due to the limited available memory and computational capability. Knowledge distillation is an effective model compression technique, which can boost the performance of a lightweight student network by transferring the knowledge from a more complex model or an ensemble of models. Due to its reduced size, this lightweight model is more suitable for deployment on edge devices. In this paper, we introduce an online knowledge distillation framework, which relies on an original attention mechanism to effectively combine the predictions of a cohort of lightweight (student) networks into a powerful ensemble, and use this as a distillation signal. The proposed aggregation strategy uses the predictions of the individual students as well as ground truth data to determine a set of weights needed for ensembling these predictions. This mechanism is solely used during system training. When testing or at inference time, a single, lightweight student is extracted and used. The extensive experiments we performed on several image classification benchmarks, both by training models from scratch (on CIFAR-10, CIFAR-100, and Tiny ImageNet datasets) and using transfer learning (on Oxford Pets and Oxford Flowers datasets), showed that the proposed framework always leads to an improvement in the accuracy of knowledge-distilled students and demonstrates the effectiveness of the proposed solution. Moreover, in the case of ResNet architecture, we observed that the knowledge-distilled model achieves a higher accuracy than a deeper, individually trained ResNet model.