If the application is normally used for the searching of an existing database of image files for the purpose of making copies or fabricating new replacement documents, it is reasonable to believe someone might make use of the text searching feature. What is not reasonable, is to believe there is any good explanation for changing pixel size and dynamic range from one character to another. The only reasonable conclusion is that the output document accurately represents the input image files.
That's to be expected when running an enhanced scan on documents that aren't uniformly clear. Some characters aren't recognized and get rendered along with the background.
For example, the same thing can be seen in this PDF: