EQFace: A Simple Explicit Quality Network for Face Recognition (2105.00634v1)

Abstract: As the deep learning makes big progresses in still-image face recognition, unconstrained video face recognition is still a challenging task due to low quality face images caused by pose, blur, occlusion, illumination etc. In this paper we propose a network for face recognition which gives an explicit and quantitative quality score at the same time when a feature vector is extracted. To our knowledge this is the first network that implements these two functions in one network online. This network is very simple by adding a quality network branch to the baseline network of face recognition. It does not require training datasets with annotated face quality labels. We evaluate this network on both still-image face datasets and video face datasets and achieve the state-of-the-art performance in many cases. This network enables a lot of applications where an explicit face quality scpre is used. We demonstrate three applications of the explicit face quality, one of which is a progressive feature aggregation scheme in online video face recognition. We design an experiment to prove the benefits of using the face quality in this application. Code will be available at \url{https://github.com/deepcam-cn/facequality}.

• The paper introduces EQFace, which computes explicit quality scores alongside feature vectors for enhanced face recognition.
• It employs an innovative loss function and iterative training pipeline to integrate quality assessment without relying on pre-annotated quality labels.
• Empirical results demonstrate state-of-the-art performance, achieving 98.18% accuracy on video face recognition benchmarks.

Analysis of "EQFace: A Simple Explicit Quality Network for Face Recognition"

The paper "EQFace: A Simple Explicit Quality Network for Face Recognition" presents a novel approach to enhance the accuracy of face recognition tasks, particularly in scenarios featuring low-quality images. The proposed method, EQFace, introduces an innovative network that simultaneously extracts feature vectors and computes an explicit, quantitative quality score for face images. This paper reports significant performance improvements over existing models, indicating its potential impact on video surveillance and other applications requiring robust face recognition capabilities.

The central contribution of EQFace lies in its ability to produce an explicit quality assessment for face images without relying on datasets with pre-annotated quality labels. The paper claims this is the first network achieving such dual functionality within the same network architecture. The introduction of a quality assessment branch to the baseline network incorporates minimal additional complexity, and the model employs a novel loss function to effectively train this branch. The face quality scores generated by the network are utilized to reweight or filter low-quality images during training and testing, demonstrating improved recognition rates in experimental evaluations across several datasets, including LFW, CALFW, CPLFW, CFPFP, YTF, IJB-B, and IJB-C.

EQFace's training pipeline is distinguished by its iterative approach, which includes three primary steps. Initially, the model is trained with a standard loss function where the quality weights are uniformly set, ensuring the effective training of feature extraction. Subsequently, the explicit quality network branch is engaged, optimizing the quality weights. In the final step, the network leverages the computed quality scores to refine the training process, potentially iterating this procedure for enhanced results. The research reports that implementing the quality-weighted face recognition (QW) approach yields noticeable improvements, particularly in the challenging video face recognition scenarios.

Empirical results indicate that EQFace achieves state-of-the-art performance on several testing datasets. Notably, on the Youtube Face Database, the accuracy reached 98.18%, which surpasses previous methods like QAN, NAN, and others. The paper also illustrates significant improvement in the 1:1 verification and 1:N identification tasks on IJB-B and IJB-C datasets, surpassing many contemporary models. This demonstrates EQFace's efficacy in handling variations caused by pose, blur, occlusion, and illumination—common challenges in unconstrained environments.

A notable aspect of EQFace is its application in real-time video recognition, where face images become available progressively. The network's ability to generate quality scores online makes it well-suited for such dynamic scenarios, reinforcing its practical relevance. The paper details a progressive feature aggregation strategy, an application of EQFace that underscores its utility in real-time surveillance applications.

The implications of this research are multifaceted. Theoretically, EQFace addresses an ongoing challenge in AI and computer vision—the assessment of image quality in training and recognition tasks. Practically, its implications extend to surveillance systems, authentication mechanisms, and any domain reliant on reliable facial recognition in dynamic and uncontrolled conditions. The research opens avenues for future work in enhancing the generalization capabilities of facial recognition systems, integrating quality assessment into multi-modal recognition tasks, and exploring the quality network's applicability in broader object recognition contexts.

In conclusion, "EQFace: A Simple Explicit Quality Network for Face Recognition" is a commendable contribution to the field, offering a reliable and efficient mechanism for incorporating image quality metrics into face recognition tasks. The simplicity and adaptability of the presented framework suggest promising future enhancements in AI-driven recognition systems.