I always expected steganography to be used to encode a compressed and encrypted datastream, not uncompressed images. Such a compressed and encrypted datastream should be indistinguishable from noise, so not show in this kind of analysis.

My pet idea for doing steganography was to use nonstandard or suboptimal encoding sequences. For compressed data formats, including images, music, video, and general files, there is almost always a choice of various encodings which will decompress to the same result. Systematically varying a compression choice gives an invisible way to encode data.