Based on the nonlinear mapping ability of kernel learning, an algorithm of kernel independent component
analysis based on wavelet kernel generalized variance(WKGV-KICA) is proposed. The wavelet kernel which is characterized
by approximate orthogonality has the advantage in local signal analysis. Related to mutual information theory, the contrast
function defined by kernel generalized variance(KGV) has desirable mathematical properties as the measure of statistical
independence. The algorithm is applied to wide-ranging blind source separation problems and compared with existing
algorithms. Experimental results show that WKGV-KICA algorithm can achieve higher separation accuracy and better
properties under the same condition.