This paper presents a study showing the benefits of the Effi-cientNet models compared with heavier Convolutional Neural Networks (CNNs) in the Document Classification task, essential problem in the digitalization process of institutions. We show in the RVL-CDIP dataset that we can improve previous results with a much lighter model and present its transfer learning capabilities on a smaller in-domain dataset such as Tobacco3482. Moreover, we present an ensemble pipeline which is able to boost solely image input by combining image model predictions with the ones generated by BERT model on extracted text by OCR. We also show that the batch size can be effectively increased without hindering its accuracy so that the training process can be sped up by parallelizing throughout multiple GPUs, decreasing the computational time needed. Lastly, we expose the training performance differences between PyTorch and Tensorflow Deep Learning frameworks.

Image morphing has proven to be very successful at deceiving facial recognition systems. Such a vulnerability can be critical when exploited in an automatic border control scenario. Recent works on this topic rely on dedicated algorithms which require additional software modules deployed alongside an existing facial recognition system. In this work, we address the problem of morphing detection by using state-of-the-art facial recognition algorithms based on hand-crafted features and deep convolutional neural networks. We show that a general-purpose face recognition system combined with a simple linear classifier can be successfully used as a morphing detector. The proposed method reuses an existing feature extraction pipeline instead of introducing additional modules. It requires neither fine-tuning nor modifications to the existing recognition system and can be trained using only a small dataset. The proposed approach achieves state-of-the-art performance on our morphing datasets using a 5-fold cross-validation.

No abstract

Facial recognition has become a critical constituent of common automatic border control gates. Despite many advances in recent years, face recognition systems remain susceptible to an ever evolving diversity of spoofing attacks. It has recently been shown that high-quality face morphing or splicing can be employed to deceive facial recognition systems in a border control scenario. Moreover, facial morphs can easily be produced by means of open source software and with minimal technical knowledge. The purpose of this work is to quantify the severeness of the problem using a large dataset of morphed face images. We employ a state-of-the-art face recognition algorithm based on deep convolutional neural networks and measure its performance on a dataset of 7260 high-quality facial morphs with varying blending factor. Using the Inception-ResNet-v1 architecture we train a deep neural model on 4 million images to obtain a validation rate of 99.96% at 0.04% false acceptance rate (FAR) on the original, unmodified images. The same model fails to repel 1.13% of all morphing attacks, accepting both the impostor and the document owner. Based on these results, we discuss the observed weaknesses and possible remedies.

This paper presents the application of evolutionary multi-objective optimization (EMO) to the improvement of a face detection system. The face detection system is based on the boosted cascade system, and analyzes image positions on different scales in a three-step-procedure. Based on threshold settings, the algorithm decides whether to continue with the test on a finer scale at the current position. Thus, the thresholds for all scales and stages have a major influence on the performance of the system, and become the subject of the evolutionary optimization according to three objectives: low false positive rate, high detection rate and low processing time. The used EMO is the extension of the Standard Genetic Algorithm to the EMO case by using Fuzzy Pareto Dominance as a meta-heuristic. The application of this EMO to the face detection system is explored and discussed using images from a standard face detection benchmark dataset. From the runtime analysis of the EMO it c an be concluded that the algorithm reliably approaches the Pareto set of the problem