Independent Component Analysis of Images
暂无分享,去创建一个
Independent component analysis (ICA) is a general machine learning method for analyzing the statistical structure of data or signals. It is based on a linear generative model with statistically independent and non-Gaussian latent variables. In computational neuroscience, it can be applied on small patches (windows) of ordinary photographic images to model their structure. The result is that the independent components are similar to the outputs of simple cells in the primary visual cortex (V1). Thus, ICA of natural images provides an interesting theory to explain why the response properties (receptive fields) of simple cells in V1 are as they are: they are adapted to the statistical structure of natural images. This supports the more general hypothesis that the visual cortex constructs an internal probabilistic model of the world and codes the incoming input efficiently.
[1] E. Oja,et al. Independent Component Analysis , 2013 .
[2] Eero P. Simoncelli,et al. Natural image statistics and neural representation. , 2001, Annual review of neuroscience.
[3] Aapo Hyvärinen,et al. Natural Image Statistics - A Probabilistic Approach to Early Computational Vision , 2009, Computational Imaging and Vision.