A collaborative deep multitask learning network for face image compliance to ISO/IEC 19794-5 standard. (15th July 2022)

• Record Type: Journal Article
• Authors: de Andrade e Silva, Arnaldo Gualberto; Gomes, Herman Martins; Batista, Leonardo Vidal
• Is Part Of: Expert systems with applications. Volume 198(2022)
• Publication Date: 2022-07-15
• Subjects: Face quality -- ICAO -- ISO/IEC 19794-5 -- Multitask Learning -- Autoencoders -- Deep Learning; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.116756
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The face is considered the primary biometric trait for machine-readable travel documents, like passports. In this context, the ISO/IEC 19794-5 standard defines a set of photographic requirements to ensure image quality and simplify the face recognition process. However, the assessment of face image compliance to the ISO/ICAO standard is still mostly performed by humans today due to the lack of automatic evaluation systems to perform this task. In this paper, we present the first deep multitask learning-based method designed for automatic evaluation of the photographic requirements of the ISO/IEC 19794-5 standard, called ICAONet. We extended undercomplete Autoencoders to employ a multi-and-collaborative learning approach, where both supervised and unsupervised learning is performed concurrently and in a collaborative manner. The method is trained using an ad hoc image dataset and evaluated by an official benchmark system also used by other approaches presented in the literature. The results show that our method achieves the best results in terms of Equal Error Rate for 9 out of the 23 photographic requirements of ISO/IEC 19794-5, which was not achieved by any other individual method evaluated. Therefore, the proposed method can be considered the best overall solution among academic works published in the literature and private SDKs. Overall, the median Equal Error Rate (3.3%) is also competitive. Finally, in terms of running time, the proposed method stands out among the fastest to evaluate all 23 requirements according to the official benchmark. Highlights: Multitask Learning can improve the results of ISO/IEC 19794-5 standard. State-of-art results in 9 out of 23 requirements across all published methods. The second best median EER compared to methods that evaluate all requirements. The fastest method to evaluate all requirements on CPU (0.15s).