Given in input a 3D digital model of a real object and a set of images from different view points, the problem considered is how to build a single texture map which integrates the color info contained in the input images, minimizing redundancy and optimizing the color attribute representation. We improve a previous solution (which stitches the images on the geometry) by introducing a new approach for a better color matching of the different adjoining texture chunks. The new solution applies cross-correlation and interpolation of the input images directly on the texture image space. Our averaging technique can remove color difference or discontinuity between adjacent sections of the 3D models which are mapped on different input images; in particular, the process is performed on the whole texture, contrary to other approaches that limit color smoothing on small transition zones.