Normalize each image channel by its supplied mean and standard deviation to produce standardized vision-model inputs.
