We propose a new POC (phase-only correlation)-based highaccuracy correspondence detection method for multi-channel images. There is the possibility of improving detection accuracy because conventional POC-based methods do not use color information. In the proposed method, a normalized cross spectrum (or cross-phase spectrum) and weight are calculated for each color channel in the Fourier domain. The weight is determined by the amplitude of the cross spectrum. The weighted normalized cross spectra of all color channels are combined and inverse Fourier transformation is conducted to obtain a POC function. An experimental evaluation of the matching accuracy between the conventional POC-based method and the proposed method shows that RMSE decreased approximately 25%. This paper also describes an application of the proposed method to a stereo image matching. The average detection ratio of correspondences between a stereo-pare image is improved 64% to 95%.