That approach is problematic with text. If you take a blurry version of the word "apple", you can't be sure if the text should be "apple" or "appie" as the "i" and "l" look too similar when blurred. Some scanners do this to get clean crisp letters and end up fixing typos or making other substitutions. "Filling in pixels" is inherently guesswork and it's certainly not "issue free".