Normalizes a scaled float image to zero mean and unit variance.

It assumes the input image is float dtype with values in [0, 1) if offset is MEAN_NORM, values in [0, 255] if offset is MEAN_RGB.

image A tf.Tensor in float32 dtype with values in range [0, 1) or [0, 255].
offset A tuple of mean values to be subtracted from the image.
scale A tuple of normalization factors.

A normalized image tensor.