cca.rar_19.comcca_CCA_CCA 相关系数_cca 相关性_相关系数

function[ccaEigvector1, ccaEigvector2] = CCA(data1, data2) % Input: % data1 ???? view1 % data2 ???? view2 % both row : a sample % column : a feature % Output: % ccaEigvector1 : the projection of view1 % ccaEigvector2 : the projection of view2 % both are not unit(length) one, it makes the conical % correlation variable has unit variance % Reference?? % Appearance models based on kernel canonical correlation analysis % Pattern Recognition, 2003 % Comments: % using SVD instead of using the eigen decomposition dataLen1 = size(data1, 2); dataLen2 = size(data2, 2); % Construct the scatter of each view and the scatter between them data = [data1 data2]; covariance = cov(data); % Sxx = covariance(1 : dataLen1, 1 : dataLen1) + eye(dataLen1) * 10^(-7) Sxx = covariance(1 : dataLen1, 1 : dataLen1); % Syy = covariance(dataLen1 + 1 : size(covariance, 2), dataLen1 + 1 : size(covariance, 2)) ... % + eye(dataLen2) * 10^(-7); Syy = covariance(dataLen1 + 1 : size(covariance, 2), dataLen1 + 1 : size(covariance, 2)); Sxy = covariance(1 : dataLen1, dataLen1 + 1 : size(covariance, 2)); % Syx = Sxy'; % using SVD to compute the projection Hx = (Sxx)^(-1/2); Hy = (Syy)^(-1/2); H = Hx * Sxy * Hy; [U, D, V] = svd(H, 'econ'); ccaEigvector1 = Hx * U; ccaEigvector2 = Hy * V; % make the canonical correlation variable has unit variance ccaEigvector1 = ccaEigvector1 * diag(diag((eye(size(ccaEigvector1, 2)) ./ sqrt(ccaEigvector1' * Sxx * ccaEigvector1)))); ccaEigvector2 = ccaEigvector2 * diag(diag((eye(size(ccaEigvector2, 2)) ./ sqrt(ccaEigvector2' * Syy * ccaEigvector2)))); end