MIT (Massachusetts Institute of Technology) Progress in Understanding Images

Abstract : The authors overall approach to the study of vision is based on a number of representations of the visible world, reviewed in previous Image Understanding Proceedings. Their work to date has concentrated primarily on the initial representations such as the primal sketch and reflectance maps, and the computation from them of depth, surface orientations, and material properties. The current emphasis is on the integration of the different sources of information, the analysis and representation of shape, the refinement and evaluation of the individual modules, the extension of their approach to deal with time varying images and moving objects, and the transfer of results to real time hardware implementation. In this report they review their recent work on the analysis of edge detection, the measurement of visual motion, the correspondence problem, the refinement and evaluation of stereo algorithms, the detection of depth discontinuities, the integration of surface maps, and the interpretation of shape from contours, and the acquisition of objects with photometric stereo.