Of course, this approach is rather out-dated now. I posted it 4 years ago. The Wolfram Language advances at an incredible speed. Here is a newer version of this:
img = ExampleData[{"TestImage", "Mandrill"}]

Then you can run:
audio = InverseSpectrogram[img] // Audio
which gives you the audio signal. Then you simply calculate the Spectorgram:
Spectrogram[audio, Frame -> False, AspectRatio -> 1, PlotRange -> All, ColorFunction -> "Rainbow"]

Cheers,
Marco