Auto-levelling or histogram equalization?

In the Quicktake for Apple II project, where we butcher low-quality 90’s era VGA images into even lower quality 256×192 grayscale pictures then dither them to 512×192, it is rather important that the images are correctly contrasted to make the end result much nicer. So far, I used histogram equalization, which is a way of spreading out evenly, the distribution of each pixel value to enhance contrast to the maximum. It has the effect of increasing contrast where there is almost none, at the expense of the original dynamic range. It works well to maximize contrast in an evenly-exposed, but a-bit-greyish image (where there are no 100% white or 100% black pixels): But on images where there are large contrasts, this has the effect of bringing darker areas to the same level of brightness as lighter areas: In the image above, you can notice that originally, the background trees are barely visible in the 8bpp grayscale buffer, and completely invisible once dithered. Once the image’s histogram…