We propose an original technique to reduce the number of colors contained in an image. This method uses the “baker’s transformation,” which obtains a statistically suitable mixture of the pixels of the image. From this mixture, we can extract several samples, which present the same characteristics as the initial image. The concept we imagine is to consider these samples as potential pallets of colors. These pallets make it possible to do an adaptive quantization of the effective number of colors. We consider, and we put in competition, three methods to obtain a single pallet. We present the baker’s transformation and we present methods to have a single pallet. The results illustrate the good visual quality reached by the quantized images. Finally, we present a comparison between our method and three classical methods of quantization.