Merging color and shape in a hierarchical pattern recognition model S Eberhardt1, C Zetzsche1, M Fahle2, K Schill1 |
|---|
1Cognitive Neuroinformatics, University of Bremen, Germany |
| When we're viewing and recognizing objects in natural scenes, shape and color information seem inseparably linked. However, neurobiological evidence suggests that color and shape processing in humans happen in a diverse number of distinct areas within the visual cortex, which provide specialized functionality for each submodality. The exact amount of parallel processing and the point of merging these information channels into a multimodal representation is still disputed. Here, we approach the problem from a computational point of view and adjust a hierarchical feed-forward pattern recognition model [Serre et al, 2007, PNAS, 104(15), 6424–6429] for color processing. We ask at which point modalities should be merged to maximize information in natural image statistics and test classification performance on natural tasks. We find that merging of color and shape should happen late in the processing hierarchy and conclude that parallel processing of submodalities rather than early merging into compound features is advantageous for efficient object recognition. |
| Up Home |
|---|