Achievement

Models of human visual perception, and computer vision algorithms, rely on the extraction of spatial and temporal frequencies as a first stage of analysis. Linear algebra based spectral analysis using higher-order matrices has proven to be a useful tool for extracting features from static and moving images. IGERT associate Edinah Gnang, working with advisor A. Elgammal, developed a new multilinear based Spectral Analysis framework, which they applied to face recognition and characterization of facial expression. Their approach rests on the analysis of higher order spatio-temporal correlations, as well as Bayesian Inference and Markov models using tensors. These models are comparable to those used by the human visual system, which is able to use the pattern of image motion generated by natural eye movements to remove correlated patterns from the image and extract essential features and shapes.