Intermediate visual representations for attentive recognition systems

摘要

Computational models of visual processes are of interest in fields such as cybernetics, robotics, computer vision and others. This thesis provides an analysis of a model of attention and of intermediate representation layers in the visual cortex that have direct impact on the next generation of object recognition strategies in computer vision. Biological inspiration - and even biological realism - is currently of great interest in the computer vision community. This thesis includes three major pieces, explained next. First, I believe that visual attention is a requirement to perform non-detection object recognition tasks. In order to test this hypothesis we compare the Selective Tuning model of attention [Tsotsos et al., 1995] to studies from psychophysics in visual search tasks involving color and 2D shapes. Second, I define a biologically plausible model of Shape Representation which incorporates intermediate layers of visual representation that have not previously been fully explored. I hypothesize that endstopping and curvature cells are of great importance for shape selectivity and show how their combination can lead to shape selective neurons. This Shape Representation model provides a highly accurate fit with neural data from [Pasupathy and Connor, 2001, Pasupathy and Connor, 2002]. Finally, in the same way curvature parts may be configured into shapes, spatial gradients of velocity vectors may be related to optic flow in a hierarchical representation of visual motion analysis. For my last contribution I provide psychophysical evidence of the role of spatial gradients of velocity in optical flow perception as well as neurophysiological evidence for neurons tuned for such gradients. Following previous authors such as [Zucker, 1981] and [Marr, 1982], I have shown that deeper understanding of visual processes in humans and non-human primates can lead to important advancements in computational perception theories and systems.