scispace - formally typeset
Search or ask a question
Proceedings ArticleDOI

Subclass Contrastive Loss for Injured Face Recognition

TL;DR: In this article, a Subclass Contrastive Loss (SCL) was proposed for face recognition in the presence of facial injuries such as swelling, bruises, blood clots, laceration, and avulsion.
Abstract: Deaths and injuries are common in road accidents, violence, and natural disaster. In such cases, one of the main tasks of responders is to retrieve the identity of the victims to reunite families and ensure proper identification of deceased/injured individuals. Apart from this, identification of unidentified dead bodies due to violence and accidents is crucial for the police investigation. In the absence of identification cards, current practices for this task include DNA profiling and dental profiling. Face is one of the most commonly used and widely accepted biometric modalities for recognition. However, face recognition is challenging in the presence of facial injuries such as swelling, bruises, blood clots, laceration, and avulsion which affect the features used in recognition. In this paper, for the first time, we address the problem of injured face recognition and propose a novel Subclass Contrastive Loss (SCL) for this task. A novel database, termed as Injured Face (IF) database, is also created to instigate research in this direction. Experimental analysis shows that the proposed loss function surpasses existing algorithm for injured face recognition.
Citations
More filters
Journal ArticleDOI
01 Jan 2021
TL;DR: In this paper, a Subclass Injured Face Identification (SCIFI) loss was proposed to learn feature representation agnostic to injury variations, which is used in learning feature representation for face recognition.
Abstract: Deaths and injuries are common in road accidents, violence, and natural disaster. In accidents and natural disasters scenarios, one of the tasks of responders is to retrieve the identity of the victims to reunite families or ensure proper identification of deceased persons. Apart from this, the identification of unidentified dead bodies due to violence and accidents is crucial for the police investigation. In the absence of identification cards, different forensic techniques such as DNA profiling and dental profiling may be used for identification. In this research, we present face recognition as a fast and viable approach for recognizing individuals with injuries. Face, which can be captured easily, is one of the most commonly used and widely accepted biometric modalities. However, face recognition is challenging in the presence of injuries as facial injuries change the appearance and geometric properties of the face due to swelling, bruises, blood clots, and accidental cuts. These changes introduce large intra-class variations among the same subject and small inter-class separability among different subjects. To address the challenge, we propose a novel Subclass Injured Face Identification (SCIFI) loss which is used in learning feature representation agnostic to injury variations. Additionally, an extended Injured Face (IF-V2) database of 150 subjects is presented to evaluate the performance of face recognition models. Multiple experiments and comparisons are performed to showcase the efficacy of the proposed SCIFI loss based face recognition.

3 citations

Journal ArticleDOI
01 Apr 2021
TL;DR: In this paper, the authors proposed a facial vector-based algorithm for filling admission forms with the help of face recognition, which could severely cut down the delays in hospital admission and treatment.
Abstract: When the person met with an accident is brought to the hospital, there are many official formalities (e.g., admission form to be filled) before the treatment that can be started. In some severe cases, these formalities can delay the treatment, which could be fatal to the patient. The automated system which can fill these forms with the help of face recognition could severely cut down the delays. But, in some cases, the injury and the blood on to the face fail facial recognition. To overcome this problem, we have proposed a facial vector-based algorithm. In the current work, we have also demonstrated, sending the SMS to the concerned authorities (police) and even to the relatives of the patient automatically using GSM modules. The patient’s information was received from centralized databases of different hospitals that are linked through the internet. We have tested the algorithm on more than 213K images from different databases like celebA, LFW, UCFI. We found that the maximum accuracy of our system was 98.23%. As a proof-of-concept, we tried testing on 51 real-time patient images and found that the accuracy is 94.11%. This automated form filling not only reduced the delay in hospital admission, but also also helped in treatment, because of the auto-filled medical history.

1 citations

Posted Content
TL;DR: This study shows that post-mortem iris recognition may be close-to-perfect approximately 5–7 h after death and occasionally is still viable even 21 days after death, which contradict the statements present in the past literature that the iris is unusable as a biometrics shortly after death.
Abstract: This paper presents a comprehensive study of post-mortem human iris recognition carried out for 1,200 near-infrared and 1,787 visible-light samples collected from 37 deceased individuals kept in the mortuary conditions. We used four independent iris recognition methods (three commercial and one academic) to analyze genuine and impostor comparison scores and check the dynamics of iris quality decay over a period of up to 814 hours after death. This study shows that post-mortem iris recognition may be close-to-perfect approximately 5 to 7 hours after death and occasionally is still viable even 21 days after death. These conclusions contradict the statements present in past literature that the iris is unusable as a biometrics shortly after death, and show that the dynamics of post-mortem changes to the iris that are important for biometric identification are more moderate than previously hypothesized. The paper contains a thorough medical commentary that helps to understand which post-mortem metamorphoses of the eye may impact the performance of automatic iris recognition. We also show that post-mortem iris recognition works equally well for images taken in near-infrared and when the red channel of visible-light sample is used. However, cross-wavelength matching presents significantly worse performance. This paper conforms to reproducible research and the database used in this study is made publicly available to facilitate research of post-mortem iris recognition. To our knowledge, this paper offers the most comprehensive evaluation of post-mortem iris recognition and the largest database of post-mortem iris images.

1 citations

Proceedings ArticleDOI
18 Jun 2021
TL;DR: In this article, the authors proposed a recognition methodology and analyzed the accuracy of different biometric methods based on deep learning strategies for a real study case, in which the considered data regard recent deaths of migrants in Mediterranean Sea.
Abstract: Forensic scientists often need to identify deceased people. The identification process mainly consists of the analysis of the DNA, dental records, and physical appearance. In humanitarian emergencies, the antemortem documentation needed for forensic analyses may be limited. In this context, face recognition plays a relevant role since antemortem pictures of missing persons are commonly made available by their families. Therefore, automatic recognition systems could be of paramount importance for reducing the search time in databases of face images and for providing a second opinion to the scientists. However, there are only few preliminary studies on automatic face recognition methods for forensic applications, and none of the works in the literature consider problems related to humanitarian emergencies. In this paper, we propose the first study on automatic face recognition for humanitarian emergencies. Specifically, we propose a recognition methodology and we analyze the accuracy of different biometric methods based on deep learning strategies for a real study case. In particular, the considered data regard recent deaths of migrants in Mediterranean Sea. The obtained results are satisfactory, and suggest that automatic recognition methods based on deep learning strategies could be effectively adopted as support tools for forensic identification.
References
More filters
Proceedings ArticleDOI
07 Jun 2015
TL;DR: A system that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure offace similarity, and achieves state-of-the-art face recognition performance using only 128-bytes perface.
Abstract: Despite significant recent advances in the field of face recognition [10, 14, 15, 17], implementing face verification and recognition efficiently at scale presents serious challenges to current approaches. In this paper we present a system, called FaceNet, that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure of face similarity. Once this space has been produced, tasks such as face recognition, verification and clustering can be easily implemented using standard techniques with FaceNet embeddings as feature vectors.

8,289 citations

Proceedings ArticleDOI
01 Jan 2015
TL;DR: It is shown how a very large scale dataset can be assembled by a combination of automation and human in the loop, and the trade off between data purity and time is discussed.
Abstract: The goal of this paper is face recognition – from either a single photograph or from a set of faces tracked in a video. Recent progress in this area has been due to two factors: (i) end to end learning for the task using a convolutional neural network (CNN), and (ii) the availability of very large scale training datasets. We make two contributions: first, we show how a very large scale dataset (2.6M images, over 2.6K people) can be assembled by a combination of automation and human in the loop, and discuss the trade off between data purity and time; second, we traverse through the complexities of deep network training and face recognition to present methods and procedures to achieve comparable state of the art results on the standard LFW and YTF face benchmarks.

5,308 citations

Proceedings ArticleDOI
17 Jun 2006
TL;DR: This work presents a method - called Dimensionality Reduction by Learning an Invariant Mapping (DrLIM) - for learning a globally coherent nonlinear function that maps the data evenly to the output manifold.
Abstract: Dimensionality reduction involves mapping a set of high dimensional input points onto a low dimensional manifold so that 'similar" points in input space are mapped to nearby points on the manifold. We present a method - called Dimensionality Reduction by Learning an Invariant Mapping (DrLIM) - for learning a globally coherent nonlinear function that maps the data evenly to the output manifold. The learning relies solely on neighborhood relationships and does not require any distancemeasure in the input space. The method can learn mappings that are invariant to certain transformations of the inputs, as is demonstrated with a number of experiments. Comparisons are made to other techniques, in particular LLE.

4,524 citations

Proceedings ArticleDOI
Qiong Cao1, Li Shen1, Weidi Xie1, Omkar M. Parkhi1, Andrew Zisserman1 
15 May 2018
TL;DR: VGGFace2 as discussed by the authors is a large-scale face dataset with 3.31 million images of 9131 subjects, with an average of 362.6 images for each subject.
Abstract: In this paper, we introduce a new large-scale face dataset named VGGFace2. The dataset contains 3.31 million images of 9131 subjects, with an average of 362.6 images for each subject. Images are downloaded from Google Image Search and have large variations in pose, age, illumination, ethnicity and profession (e.g. actors, athletes, politicians). The dataset was collected with three goals in mind: (i) to have both a large number of identities and also a large number of images for each identity; (ii) to cover a large range of pose, age and ethnicity; and (iii) to minimise the label noise. We describe how the dataset was collected, in particular the automated and manual filtering stages to ensure a high accuracy for the images of each identity. To assess face recognition performance using the new dataset, we train ResNet-50 (with and without Squeeze-and-Excitation blocks) Convolutional Neural Networks on VGGFace2, on MS-Celeb-1M, and on their union, and show that training on VGGFace2 leads to improved recognition performance over pose and age. Finally, using the models trained on these datasets, we demonstrate state-of-the-art performance on the IJB-A and IJB-B face recognition benchmarks, exceeding the previous state-of-the-art by a large margin. The dataset and models are publicly available.

2,365 citations

Journal ArticleDOI
TL;DR: Experimental results show that the proposed framework can utilize large-scale noisy data to learn a Light model that is efficient in computational costs and storage spaces and achieves state-of-the-art results on various face benchmarks without fine-tuning.
Abstract: The volume of convolutional neural network (CNN) models proposed for face recognition has been continuously growing larger to better fit the large amount of training data. When training data are obtained from the Internet, the labels are likely to be ambiguous and inaccurate. This paper presents a Light CNN framework to learn a compact embedding on the large-scale face data with massive noisy labels. First, we introduce a variation of maxout activation, called max-feature-map (MFM), into each convolutional layer of CNN. Different from maxout activation that uses many feature maps to linearly approximate an arbitrary convex activation function, MFM does so via a competitive relationship. MFM can not only separate noisy and informative signals but also play the role of feature selection between two feature maps. Second, three networks are carefully designed to obtain better performance, meanwhile, reducing the number of parameters and computational costs. Finally, a semantic bootstrapping method is proposed to make the prediction of the networks more consistent with noisy labels. Experimental results show that the proposed framework can utilize large-scale noisy data to learn a Light model that is efficient in computational costs and storage spaces. The learned single network with a 256-D representation achieves state-of-the-art results on various face benchmarks without fine-tuning.

617 citations