Abstract: The watermark consists of a pseudorandom binary sequence that is embedded imperceptibly to the discrete wavelet transform coefficients of the three largest detail subbands of each color channel according to the masking and color perception characteristics of human visual system (HVS). Watermark detection uses the global correlation value compared to a threshold to decide whether or not the image contains watermarks by taking into account the cross-correlation between color bands. An optimum threshold is theoretically set to minimize the probability of missing the watermark subject to a given false detection rate (Neyman-Pearson criterion). The new threshold selection mechanism depends only on the watermarked image, and can thus be computed a posteriori, without any knowledge about the watermark energy. Experimental results demonstrate the validity of new approach.