|Every day DeepLearning and CNN results are advancing the state of the art in the area of image understanding, image processing and computer vision.
In our lab we perform cover many aspects of this evolving area.
- Convolutional Neural Network for Intermediate View Enhancement in Multiview Streaming, IEEE Trans. on Multimedia 2018
- Post-synaptic potential regularization has potential, presented at International Conference on Artificial Neural Networks 2019
Gait recognition and biometrics
|Biometrics is the science that studies the human characteristics for anthropometry research, people identification, access control and many more. Biometric features are measurable data classified as physical or behavioral. The former are related to the body and its shape. Some examples are face, hand, iris, retina and fingerprint. Behavioral characteristics are associated to particular human action, for instance handwriting and walking. Automatic recognition systems are often expensive, intrusive and require the cooperation of the subject during the acquisition. The latter cannot be always guaranteed, for instance in a video surveillance context. In this case, it iuseful to recognize people through biometric parameters that can be captured at a distance and without the collaboration of the person, such as gait. Gait analysis finds interest not only in video surveillance systems and forensics science. Many applications analyze the gait in order to discover pathologies of the body movement, rehabilitation therapy, identify the fall risk in elderly population in order to assess the frailty syndrome. All these applications are based on the analysis of video and 2D images. Images and videos are processed in order to collect gait parameters applying both model-based approaches, using the definition of a 3D model of the body in movement, or by model-free approaches, that process the silhouette of a walking person.|
- Gait Recognition
3D Video Coding and Representation
| During the last decade 3D-Cinema and 3D-television have become increasingly popular due to its superior quality of experience achieved through stereoscopy and autostereoscopy. The depth sensation in 3D television is achieved by rendering at least two views of the same scene from two slightly different viewpoints. It is desirable to use as few views as possible to achieve maximum parallax in 3D vision and it poses new challenges for coding multiviews and generating virtual views from available set of view views.
Our group is working on three major areas of 3DTV: 3D video representation & coding, virtual view generation & quality enhancement, and quality assessment of 3D videos. Selected contributions in this area are presented in the following.
- 3D Video Coding
- Panorama view with spatiotemporal occlusion compensation for 3D video coding
- A panoramic 3D video coding with directional depth aided inpainting
- 3D View Synthesis & Visual Quality Enhancement
- Edge enhancement of depth based rendered image
- Depth Image Based Rendering with Inverse Mapping (VSIM)
- Edges shape enforcement for visual enhancement of depth image based rendering
- 3D Quality Assessment
| A lot of research results in the fields of multimedia signal processing, networking and digital communications have recently contributed to the advent of our current multimedia enriched communication experience. In this area EIDOS expertise includes image and video coding, video error concealment, error resilient video coding, joint source channel coding, multiple description coding and distributed source coding, distributed and P2P video streaming, network coding.
- A. Fiandrotti, V. Bioglio, M. Grangetto, R. Gaeta, E. Magli "Band Codes for Energy-Efficient Network Coding with Application to P2P Mobile Streaming", IEEE TRANSACTIONS ON MULTIMEDIA, 2015
- R. Gaeta, M.Grangetto "DIP: Distributed identification of polluters in P2P live streaming", ACM TRANSACTIONS ON MULTIMEDIA COMPUTING, COMMUNICATIONS AND APPLICATIONS, 2014
- Enrico Baccaglini, M. Grangetto, E. Quacchio, and S. Zezza. A study of an hybrid cdn–p2p system over the planet lab network, SIGNAL PROCESSING-IMAGE COMMUNICATION, 2012
- A. Magnetto, R. Gaeta, M. Grangetto, and M. Sereno. Turinstream: A totally push, robust and efficient p2p video streaming architecture. IEEE TRANSACTIONS ON MULTIMEDIA, 2010
- M. Grangetto, E. Magli, and G. Olmo. Distributed arithmetic coding for the slepian-wolf problem. IEEE TRANSACTIONS ON SIGNAL PROCESSING,
- M. Grangetto,B. Scanavino, G. Olmo, and S. Benedetto. Iterative decoding of serially concatenated arithmetic and channel codes with JPEG 2000 applications. IEEE TRANSACTIONS ON IMAGE PROCESSING, 2007
- M. Grangetto, P. Cosman and G. Olmo, Joint source/channel coding and map decoding of arithmetic codes. IEEE TRANSACTIONS ON COMMUNICATIONS, 2005.
Biomedical Image Processing
|Radiography is nowadays a common medical exam, used for diagnosing several diseases, but has the disadvantage of exposing people to a dose of radiation. For this reason, it is important to study methods for reducing such dose. To this end we are currently developing the ViRiS - RX Simulator|
|Computer Vision and Deep Learning have tremendous potential in the areas of Biomedical Image Processing and Computer Aided Diagnosis systems. In this field we perform research thanks to the collaboration of the medical departments of our university and European and industrial partners.|
In the biomedical area we are also working on Segmentation of tumors in PET images in collaboration with the nuclear medicine group of IRCC in Candiolo . Our research results are contributing to devise automatic techniques to trim tumoral regions in dynamic PET images.
See also our book on RX imaging (in italian) L'immagine digitale in diagnostica per immagini
- Global and local anomaly detectors for tumor segmentation in dynamic pet acquisitions
- Automatic method for tumor segmentation from 3-points dynamic PET acquisitions
- Automatic GTV contouring applying anomaly detection algorithm on dynamic FDG PET images
- A GPU Simulation Tool for Training and Optimisation in 2D Digital X-Ray Imaging
The diffusion of digital media and the consequent necessity of protection has posed new challenges and problems in the computer science field. At the same time objects of large dimensions (like images, sounds or videos) and containing a large quantity of redundant data allow us for compression or, alternatively, storage of extra information. In the field of data hiding many different techniques have been developed. In EIDOs we are working on application fragile reversible watermarking capable to reveal modification of the digital object, possibly localizing the modified area with particular attention to biomedical imaging data.
- D. Cavagnino, M. Lucenteforte, M.Grangetto "High capacity reversible data hiding and content protection for radiographic images", Elsevier SignalProcessing, 2015.
Image processing for the Holy Shroud
The Eidos group has developed a long lasting experience, since the lab foundation, in the study of the images of the Holy Shroud kept in the Turin Cathedral.
Our current research in this area is in collaboration with the Collaborative System Engineering Centre in Thales Alenia Space. Our efforts are aimed to study new technologies that may be used in the robotic or human space exploration, with particular attention to Mars exploration. In this context we participate to the development of virtual simulations of space modules and environments.