SIBGRAPI 2018 Tutorials
The large variety of medical image modalities (e.g. Computed Tomography, Magnetic Resonance Imaging, and Positron Emission Tomography) acquired from the same body region of a patient together with recent advances in computer architectures with faster and larger CPUs and GPUs allows a new, exciting, and unexplored world for image registration area. A precise and accurate registration of images makes possible understanding the etiology of diseases, improving surgery planning and execution, detecting otherwise unnoticed health problem signals, and mapping functionalities of the brain. The objective of this tutorial is to present the state-of-the-art in medical image registration starting from the preprocessing steps, covering the most popular methodologies of the literature and finish with the more recent advances and perspectives from the application of Deep Learning architectures.
Face recognition made tremendous leaps in the last five years with a myriad of systems proposing novel techniques substantially backed by deep convolutional neural networks (DCNN). Though face recognition performance skyrocketed using deep-learning, leading to the belief that this technique reached human performance, yet unconstrained face recognition remains an open problem. As deep-learning allowed achieving nearly perfect accuracy on the LFW dataset, the newly released IJB sets showed how face recognition remains a difficult problem in unconstrained environments. This tutorial summarizes the main advances in deep face recognition and, more in general, in learning face representations for verification and identification. The tutorial provides a clear, structured presentation of the principal, state-of-the-art face recognition techniques appeared in the last five years in top computer vision venues. The participants will be guided firstly to understand the face recognition problem and its evaluation criteria (closed-set or open-set identification; verification) and, then, to hear about very recent methods for deep face recognition.
The number of visual surveillance systems deployed worldwide has been growing continuously. In recent years, several attempts have been made to increase the levels of automated analysis of such systems and reliably recognize human beings in fully covert conditions. Among other choices, one interesting possibility is to make use of master-slave architectures to acquire high resolution data of subjects heads and perform biometric recognition on such data. According to this paradigm, this tutorial will provide a comprehensive overview of the main phases behind the development of recognition system working in a surveillance scenario, describing frameworks/methods to: 1) use coupled visual surveillance and Pan-Tilt-Zoom (PTZ) imaging devices in surveillance settings, with a wide-view camera covering the whole scene, and a synchronized PTZ device collecting high-resolution; 2) use soft biometric information (e.g., body metrology and gait) for indexing/retrieval identities and prune the set of potential enrolled identities for each query; 3) use strong biometric traits (face) for watch list detection; and 4) faithfully balance ethics/privacy and safety/security issues in this kind of systems.