Face Detection Project
Face detection is the first fundamental step for any real application which involves face analysis. Our method uses an Adaboost classifier and an eye detector based on the Support Vector Machines
Face Recognition Project
We have developed a local-based face recognition system whose goal is to recognize a person referring to a closed gallery. Given a face image, the method starts localizing a set of fiducial points in correspondence to the facial features; afterwards it characterizes the face applying a bank of multi-resolution and multi-orientation Gabor filters at the portions of the image surrounding the fiducial points.
The system is scale-independent, robust to partial occlusions and different poses or illuminations, while we are working to make it more robust to expression variations.
3D Face Reconstruction Project
Aim of this project is to reconstruct the 3D model of a face from two or more stereo-images: at first the facial fiducial points are determined on each image and their 3D positions are estimated; the 3D model is obtained adapting a face standard model to the 3D fiducial points, and automatically applying the texture to it.
Image processing applied in industrial environment
We have designed and implemented algorithms to solve some problems arising in industrial plants. (In collaboration with CESI)
Computer Aided Diagnosis (CAD) Systems: automatic detection of lung nodules on digital X-ray chest radiographs.
Automatic Segmentation of Abdominal Organs: automatic segmentation of abdominal organs from computed tomography (CT) images.
Biometric Analysis of fetal Brain: analysis of magnetic resonance images (MRI) of fetal brain to derive salient biometric features signalizing abnormalities.
Automatic Segmentation of Reporter Mice: automatic segmentation of bioluminescent and fluorescent mouse images to automatically measure the drug effects.
Analysis of Gel 2D Images: automatic comparison of Gel 2D images.
Improvements in Machine Learning Field: development of novel techniques founded on Tensor Voting.
Development of Fisher-based classification techniques:
Development of novel classifiers coping with the Small Sample Size problem by employing a Fisher Subspace estimation.