We present a method to accurately estimate dense motion vectors between two successive color images in a sequence. This study integrates the red, green and blue channels of color images and extends the previous method³⁰ in which the Cai–Wang wavelet representation was employed as a hierarchical motion model to match gray scaled images. The coarse-to-fine wavelet-based motion model is substituted into the objective function where the sum of squared differences (SSD) between two color images is minimized iteratively. Compared with the gray scaled images, our simulated experiments show that color images have better trackability and smaller condition numbers during the optimization process, leading to faster convergence rate and more accurate results. The estimated dense correspondences have been applied effectively to create realistic novel views using trilinear constraints.