Computational model of the second-order visual channels
Denis V. Yavna, Vitaly V. Babenko · 2016
#Original presentation contains animated GIFs #Please email me if you want to get the original file (OpenDocument format) The aim of our research is to create the computational model of visual second-order channels which independently detect spatial modulations of contrast, orientation, or spatial frequency in static visual scenes. The reason to create the model is the psychophysical and psychophysiological data that support an assumption of the human second-order channel specificity. The advantage of the model is similarity with natural neural structures. Our model is based on the classical "filter-rectify-filter" scheme, which explains the second-order feature detection. In order to provide the specificity to modulation dimension, the contrast normalization mechanism was added to channels detecting the orientation and frequency modulations [Kingdom at al., 2003], and the input of inhibitory subfields of the second order mechanism was altered [Babenko, Yavna, 2009]. While the excitatory subfield is formed by the inputs from the first-order filters tuned to the certain orientation and spatial frequency band, inhibitory subfields receive signals from the elements with different orientation and frequency tunings in our model. The model can be useful for solving some of the image and video processing tasks such as segmentation, feature extraction, and data compression.