K.U.Leuven > ESAT > PSI > Visics > Research > Topics > Item 2.1

Variability in Human Shape and Speech (GOA VHS+)

L. Van Gool, J. Vanden Wyngaerd

The three groups within PSI contribute to the project: the medical group MIC, the group dealing with vision of industry, communication, and services VISICS, and the speech recognition group. The project investigates ways to robustly describe and recognize aspects of human body shape, as well as human speech. The project combines expertise in areas such as invariant-based shape descriptions, the compact and effective parameterisation of shape variability, language models to cater for different applications and different groups of speakers, etc. It focuses on aspects such as shape comparison and registration for rigid as well as non-rigid shapes, on the segmentation of shapes driven by models which take natural shape variability into account, the identification and recognition of shapes (e.g., faces) under changing viewing conditions, the interpretation of (speech) signals in noisy circumstances, etc. The project will yield tools for medical image processing, forensics, animation, etc.

The figure above shows on the right the combined image of a human face after comparison and registration, starting from the two images on the left.