Efficient Coding for Natural Images Based on the Sparseness of Neural Coding in V1 Across the Stimuli

LZ Liao,SW Luo,LW Zhao,M Tan
DOI: https://doi.org/10.1109/icnnb.2005.1614710
2005-01-01
Abstract:The sparse coding and independent component analysis for natural scenes, in recent years, have succeeded in for finding a set of basis functions that can effectively represent the input data, by supposing that the feature vectors of images should be sparse or independent. In this paper, we investigated the efficient coding for natural images by making assumptions of sparseness and independence on the activities of basis functions over the image ensemble, without considering directly the statistics of the feature vectors of images. Experimental results show that the approach can also produce basis functions which have similar properties with the receptive fields of simple cells in V1 and thereby be effective.
What problem does this paper attempt to address?