OCR is notorious for not working right, unless you know the fonts being used before hand. Also throw on top of that the Text is on an image, that further screws up software detection.
What I might do is just try a few, and look at both the png and jpg versions and see which is more pleasant when trying read the text. Maybe the fonts are big enough that it dosn't matter much. Jpgs are definitely smaller, and even simple pngs can be quite large. Its a size versus clarity tradeoff.
If you have the time and inclination, you might approach this scientifically, and convert a series of bmp images to corresponding jpg's and png's. Then use your edge detection software to compare their overlays. That ought to be proof enough of the exact differences. Doing it with PDL might be the easiest way to do that comparison.