Recent advancements in face recognition (FR) technology in surveillance systems make it possible to monitor a person as they move around. FR gathers a lot of information depending on the quantity and data sources. The most severe privacy concern with FR technology is its use to identify people in real-time public monitoring applications or via an aggregation of datasets without their consent. Due to the importance of private data leakage in the FR environment, academia and business have given it a lot of attention, leading to the creation of several research initiatives meant to solve the corresponding challenges. As a result, this study aims to look at privacy-preserving face recognition (PPFR) methods. We propose a detailed and systematic study of the PPFR based on our suggested six-level framework. Along with all the levels, more emphasis is given to the processing of face images as it is more crucial for FR technology. We explore the privacy leakage issues and offer an up-to-date and thorough summary of current research trends in the FR system from six perspectives. We also encourage additional research initiatives in this promising area for further investigation.

License Plate Character Recognition (LPCR) is a technology for reading vehicle registration plates using optical character recognition from images and videos, and it has a long history due to its usefulness. While LPCR has been significantly improved with the advance of deep learning, training deep networks for LPCR module requires a large number of license plate (LP) images and their annotations. Unlike other public datasets of vehicle information, each LP has a unique combination of characters and numbers depending on the country or the region. Therefore, collecting a sufficient number of LP images is extremely difficult for normal research. In this paper, we propose LP-GAN, an LP image generation method, by applying an ensemble of generative adversarial networks (GAN), and we also propose a modified lightweight YOLOv2 model for an efficient end-to-end LPCR module. With only 159 real LP images available online, thousands of synthetic LP images were generated by using LP-GAN. The generated images not only looked similar to real ones, but they were also shown to be effective for training the LPCR module. As a result of performance tests with 22,117 real LP images, the LPCR module trained with only the generated synthetic dataset achieved 98.72% overall accuracy, which is comparable to that of training with a real LP image dataset. In addition, we improved the processing speed of LPCR about 1.7 times faster than that of the original YOLOv2 model by using the proposed lightweight model.

Videofluoroscopic swallowing study (VFSS) is a standard diagnostic tool for dysphagia. To detect the presence of aspiration during a swallow, a manual search is commonly used to mark the time intervals of the pharyngeal phase on the corresponding VFSS image. In this study, we present a novel approach that uses 3D convolutional networks to detect the pharyngeal phase in raw VFSS videos without manual annotations. For efficient collection of training data, we propose a cascade framework which no longer requires time intervals of the swallowing process nor the manual marking of anatomical positions for detection. For video classification, we applied the inflated 3D convolutional network (I3D), one of the state-of-the-art network for action classification, as a baseline architecture. We also present a modified 3D convolutional network architecture that is derived from the baseline I3D architecture. The classification and detection performance of these two architectures were evaluated for comparison. The experimental results show that the proposed model outperformed the baseline I3D model in the condition where both models are trained with random weights. We conclude that the proposed method greatly reduces the examination time of the VFSS images with a low miss rate.

Face privacy concerns revolve around the ethical, social, and technological implications of collecting, storing, and using facial data. With the advancement of deep learning techniques, realistic face privacy involves techniques that obscure or alter facial features effectively without compromising the usability or quality of the visual content. Modern face privacy techniques suffer from three main problems: (1) lack of human perception, (2) indistinguishability, and (3) loss of facial attributes. Modern face privacy techniques generate random, realistic faces to conceal the identifiable features of the original faces but lack the application of human perception to face de-identification. Indistinguishability arises with the highly realistic nature of fake faces used in face privacy, making it difficult to distinguish whether a face has been manipulated or not. Most face-privacy methods fail to retain the facial attributes of the de-identified faces. Our face de-identification method is designed to address all three issues mentioned. We propose a novel face de-identification method that considers both human perception and face recognition models when de-identifying a face. We explore the tradeoff between a user misidentifying the original identity with a well-known celebrity and a facial recognition model that tries to identify the original identity. We generate caricature faces of the de-identified faces to ensure our manipulated faces can be distinguished effortlessly. The face caricatures are the exaggeration of the eyes and mouth region, and we provide different exaggeration scales depending on preference and application. We perform an attribute preservation optimization process to retrieve all the facial attributes. We demonstrate our method through a series of both qualitative and quantitative experiments with numerous user studies.

This study concentrates on machine learning, specifically deep learning techniques, to automatically detect the presence of aspiration or penetration in videofluoroscopic swallowing studies (VFSS). A comparative analysis is conducted on various deep learning architectures such as 2D Convolutional Neural Networks (2D-CNN), Long Short-Term Memory (LSTM), and 3D Convolutional Neural Networks (3D-CNN). This comparison assesses the performance, network size, and computational speed of the models. In addition, we present findings derived from multi-label and multi-class classification methods. By evaluating the strengths and weaknesses of each technique, we propose the most effective method for detecting penetration or aspiration in VFSS. Our comprehensive evaluation reveals the superiority of 3D-CNN in the automatic detection of penetration and aspiration in VFSS. This research contributes to the development of a clinically viable automatic detection system, offering potential advancements in the care and management of patients with dysphagia.