Ester Gonzalez-Sosa

Image-based gender estimation from body and face across distances

Gender estimation has received increased attention due to its use in a number of pertinent security and commercial applications. Automated gender estimation algorithms are mainly based on extracting representative features from face images. In this work we study gender estimation based on information deduced jointly from face and body, extracted from single-shot images. The approach addresses challenging settings such as low-resolution-images, as well as settings when faces are occluded. Specifically the face-based features include local binary patterns (LBP) and scale-invariant feature transform (SIFT) features, projected into a PCA space. The features of the novel body-based algorithm proposed in this work include continuous shape information extracted from body silhouettes and texture information retained by HOG descriptors. Support Vector Machines (SVMs) are used for classification for body and face features. We conduct experiments on images extracted from video-sequences of the Multi-Biometric Tunnel database, emphasizing on three distance-settings: close, medium and far, ranging from full body exposure (far setting) to head and shoulders exposure (close setting). The experiments suggest that while face-based gender estimation performs best in the close-distance-setting, body-based gender estimation performs best when a large part of the body is visible. Finally we present two score-level-fusion schemes of face and body-based features, outperforming the two individual modalities in most cases.

Comparison of Body Shape Descriptors for Biometric Recognition Using MMW Images

The use of Millimetre wave images has been proposed recently in the biometric field to overcome certain limitations when using images acquired at visible frequencies. In this paper, several body shape-based techniques were applied to model the silhouette of images of people acquired at 94 GHz. We put forward several methods for the parameterization and classification stage with the objective of finding the best configuration in terms of biometric recognition performance. Contour coordinates, shape contexts, Fourier descriptors and silhouette landmarks were used as feature approaches and for classification we utilized Euclidean distance and a dynamic programming method. Results showed that the dynamic programming algorithm improved the performance of the system with respect to the baseline Euclidean distance and the necessity of a minimum resolution of the contour to achieve promising equal error rates. The use of the contour coordinates is the most suitable feature to use in the system regarding the performance and the computational cost involved when having at least 3 images for model training. Besides, Fourier descriptors are more robust against rotations, which may be of interest when dealing with few training images.

Body shape-based biometric recognition using millimeter wave images

The use of MMW images has been proposed recently in the biometric field aiming to overcome certain limitations when using images acquired at visible frequencies. In this paper, several body shape-based techniques are applied to model the silhouette of images of people acquired at 94 GHz. Three main approaches are presented: a baseline system based on the Euclidean distance, a dynamic programming method and a procedure using Shape Contexts descriptors. Results show that the dynamic time warping algorithm achieves the best results regarding the system performance (around 1.3% EER) and the computation cost. Results achieved here are also compared to previous works based on the extraction of geometric measures between several key points of the body contour. An average relative improvement of 33% EER is achieved for the work reported here.

Millimetre wave person recognition: Hand-crafted vs learned features

Imaging using millimeter waves (mmWs) has many advantages including ability to penetrate obscurants such as clothes and polymers. Although conceal weapon detection has been the predominant mmW imaging application, in this paper, we aim to gain some insight about the potential of using mmW images for person recognition. We report experimental results using the mmW TNO database consisting of 50 individuals based on both hand-crafted and learned features from Alexnet and VGG-face pretrained CNN models. Results suggest that: i) mmW torso region is more discriminative than mmW face and the entire body, ii) CNN features produce better results compared to hand-crafted features on mmW faces and the entire body, and iii) hand-crafted features slightly outperform CNN features on mmW torso.

Dealing with occlusions in face recognition by region-based fusion

The last research efforts made in the face recognition community have been focusing in improving the robustness of systems under different variability conditions like change of pose, expression, illumination, low resolution and occlusions. Occlusions are also a manner of evading identification, which is commonly used when committing crimes or thefts. In this work we propose an approach based on the fusion of non occluded facial regions that is robust to occlusions in a simple and effective manner. We evaluate the region-based approach in three face recognition systems: Face++ (a commercial software based on CNN) and two advancements over LBP systems, one considering multiple scales and other considering a larger number of facial regions. We report experiments based on the ARFace database and prove the robustness of using only non-occluded facial regions, the effectiveness of a large number of regions and the limitations of the commercial system when dealing with occlusions.

Pose Variability Compensation Using Projective Transformation for Forensic Face Recognition

The forensic scenario is a very challenging problem within the face recognition community. The verification problem in this case typically implies the comparison between a high quality controlled image against a low quality image extracted from a close circuit television (CCTV). One of the downsides that frequently presents this scenario is pose deviation since CCTV devices are usually placed in ceilings and the subject normally walks facing forward. This paper proves the value of the projective transformation as a simple tool to compensate the pose distortion present in surveillance images in forensic scenarios. We evaluate the influence of this projective transformation over a baseline system based on principal component analysis and support vector machines (PCA-SVM) for the SCface database. The application of this technique improves greatly the performance, being this improvement more striking with closer images. Results suggest the convenience of this transformation within the preprocessing stage of all CCTV images. The average relative improvement reached with this method is around 30% of EER.

Exploring automatic extraction of body-based soft biometrics

Given the growing interest in soft biometrics and its application in many areas related to biometrics, this paper focuses on the automatic extraction of body-based soft biometric attributes from single-shot images. The selected body soft biometrics are: height, shoulder width, hips width, arms length, body complexion and hair colour. For the extraction of these attributes, the Southampton Multi-Biometric Tunnel Database has been used with a total of 222 subjects. Images at far distance between the subject and the camera were considered in order to be able to extract the whole body of the person. Feature extraction is based on distances between key points automatically extracted from the persons silhouette, and also based on pixel information. Support Vector Machines (SVM) are used as the matchers, achieving promising results. Finally, given an image of a person at a distance, the system automatically gives the probability for the classes of each body-based soft biometrics considered, which could be seen as a description of the subjects body. This description could be used to reduce the search space in forensic applications, or to improve the robustness of biometric recognition systems at a distance, especially for face and gait systems, among other applications.

Body Shape-Based Biometric Person Recognition from mmW Images

A growing interest has arisen in the security community for the use of millimeter waves in order to detect weapons and concealed objects. Also, the use of millimetre wave images has been proposed recently for biometric person recognition to overcome certain limitations of images acquired at visible frequencies. This paper proposes a biometric person recognition system based on shape information extracted from millimetre wave images. To this aim, we report experimental results using millimeter wave images with different body shape-based feature approaches: contour coordinates, shape contexts, Fourier descriptors and row and column profiles, using Dynamic Time Warping for matching. Results suggest the potential of performing person recognition through millimetre waves using only shape information, a functionality that could be easily integrated in the security scanners deployed in airports.

Exploring Body Shape From mmW Images for Person Recognition

Due to the ability of millimeter waves (mmWs) to penetrate dielectric materials, such as plastic, polymer, and clothes, the mmW imaging technology has been widely used for the detection of concealed weapons and objects. The use of mmW images has also recently been proposed for biometric person recognition to overcome certain limitations in image acquisition at visible frequencies. This paper proposes a biometric person recognition system based on the shape information extracted from real mmW images. To this aim, we report experimental results using the mmW images with different body shape-based feature approaches, such as contour coordinates, shape contexts, Fourier descriptors, and row and column profiles. We also study various distance-based and classifier-based matching schemes. Experimental results suggest the potential of performing person recognition through mmW imaging using only shape information, a functionality that could be integrated in the security scanners deployed in airports.