Abstract
We propose to assess the relevance of theories of synaptic modification as models of feature extraction in human vision, by using masks derived from synaptic weight patterns to occlude parts of the stimulus images in psychophysical experiments. In the experiment, we found that a mask derived from principal component analysis of object images was more effective in reducing the generalization performance of human subjects than a mask derived from another method of feature extraction (BCM), based on higher-order statistics of the images.