The proposed scanned graphics palette extraction process starts with the extraction of colors of uniform regions, then the colors of local features. In order to enhance the discrimination of unsaturated colors, the image is converted into L*a*b* space and the CMC color difference is used together with the Euclidean distance. After the palette extraction, a regularization process modifies the color of a pixel in order to make it coherent with its neighborhood, thus reducing the number of connected components, hence providing one step further to vectorization. Results are shown on scanned maps and on some scanned graphics.