The rapidly growing demand of 3D holographic imaging systems has created a significant interest in many disciplines. However, these scanned images reveal human body details and raise privacy concerns. In this project, the team develops: 1) a virtual test bed for privacy-aware imaging. Similar to the "phantoms" widely used in medical studies, the team intends to develop a device-independent test database that can simulate the output of a 3D human body scanner for a broader research community. 2) The team develops robust and fast algorithms for segmenting human bodies and detecting private body parts. Many machine learning algorithms are coordinate-dependent and limited by the training data space. Existing algorithms only work within small bounding boxes that do not warrant an acceptable performance. In this project, the team uses innovative Analogia Graph and Relative Template Matching algorithms for speed and accuracy. 3) The team develops the algorithm for detecting anomalous objects on human body using curvature measurement and spatial density filtering models. Finally, 4) the team develops a method for assessing the visual privacy algorithms based on detection accuracy, privacy and aesthetic measurements.
The anticipated results will have broader impacts on privacy-aware security in airports and other large infrastructures. It can benefit custom-fit products that are designed from personal 3D scanning data. It can be used in reconstruction of ancient artifacts in digital archeology. In addition, it can be applied to medical diagnoses, such as virtual colonoscopy. Finally, this project can benefit multidisciplinary education of human-centric systems.
