The way images work is that they only have coordinates for the corners. These get transformed and then the graphics hardware fills in the interior of the image using the texturemapping hardware. The texture lookup is linear (actually the ratio of two linear interpolations, but that doesn't matter here), so making one of the scales log has no effect on the interior of the image.
To get the interior of the image to transform non-linearly, you need coordinates for the interior vertices, rather than just colors. The pcolor command is the simplest way to do this:
h = pcolor(x,y,log10(image_spectrogram(1:floor(Fs/2),:)+1));
h.EdgeColor = 'none';
What pcolor is doing is actually creating a surface object and setting the view so that you're looking down from the top. It is important to note that the surface object is going to consume more memory than the image object would. That's because it actually has coordinates for all of the interior vertices.
