Natural Image Statistics A probabilistic approach to early computational vision 自然图像统计早期计算视觉的一种概率方法 源代码.doc

readme This directory contains the Matlab(TM) code and natural image data to reproduce most of the experiments on in chapters 1-11 and Section 13.1 of the book "Natural Image Statistics" by Hyvarinen, Hurri, and Hoyer, published by Springer. Code by Aapo Hyvarinen partly based on code by Patrik O. Hoyer and rotavg.m code courtesy of Bruno Olshausen This is version 1.0, 11 Dec 2008, run with Matlab 7.5.0 MAIN FILE: figures.m Running this file will do all the main experiments DATA DIRECTORY: data/ This directory contains the natural images used in the experiments SAMPLING: sampleimages.m Sample patches from images in directory "data" PREPROCESSING: removeDC.m Remove DC from image patches Note: there is no single command for whitening the data, see figures.m on how to use the results from pca.m to whiten your data MAIN ALGORITHMS: pca.m %principal component analysis ica.m %FastICA for whitened data isa_est.m %estimation of ISA model for whitened data tica.m %estimation of topographic ICA for whitened data overcomplete.m %estimation of energy-based overcomplete model for whitened data OTHER FILES: subroutines of the above, including plotting and terminal reporting commands. removedc % Removes DC component from image patches % Data given as a matrix where each patch is one column vectors % That is, the patches are vectorized. function Y=removeDC(X); % Subtract local mean gray-scale value from each patch in X to give output Y Y = X-ones(size(X,1),1)*mean(X); return; rotavg % rotavg.m - function to compute rotational average of (square) array % by Bruno Olshausen % % function f = rotavg(array) % % array can be of dimensions N x N x M, in which case f is of % dimension NxM. N should be even. % function f = rotavg(array) [N N M]=size(array); [X Y]=meshgrid(-N/2:N/2-1,-N/2:N/2-1); [theta rho]=cart2pol(X,Y); rho=round(rho); i=cell(N/2+1,1); for r=0:N/2 i{r+1}=find(rho==r); end f=zeros(N/2+1,M); for m=1:M a=array(:,:,m); for r=0:N/2 f(r+1,m)=mean(a(i{r+1})); end end