| The pressures of evolution have maximized the brain's processing speed and metabolic efficiency. Therefore, to understand why neurons in early visual cortex respond to specific visual patterns, you simply have to learn an appropriate efficient code. Here, we used independent components analysis (ICA) on a series of whitened image patches taken from pictures of natural scenes - rocks, trees, fields... For each filter, the "neuron" will fire if the image is bright on the bright spots and dark on the dark spots (more precisely, it's a linear filter). Some of these resemble simple cells in early visual cortex, as they have a characteristic oriented bright/dark pattern (like 2-D gabor functions, to be more precise). The idea is that the more filters derived from an efficient coding of natural scenes resemble those of neurons in early visual cortex, the better we understand how the brain processes visual input. For a more thorough understanding please visit http://emva.net/educational/introduction.html
Contributed by: Mark V. Albert
|
|