Add --nopictures, --tables=n to cli

[Ben Jackson]

I found that cuneiform would not OCR anything inside an outline (any kind of box) and would consider it either a picture or a table. With '--nopicture' it seems to just ignore those same areas. If you also add --tables it will successfully OCR inside the table.

I've never used bzr before this so I've probably botched the patch somehow, but it is fairly trivial.

[Ben Jackson (ben.jackson)]

I see now in puma.h there are constants related to the values I set:

        # define PUMA_TABLE_NONE 0
        # define PUMA_TABLE_DEFAULT 1
        # define PUMA_TABLE_ONLY_LINE 2
        # define PUMA_TABLE_ONLY_TEXT 3
        # define PUMA_TABLE_LINE_TEXT 4

        # define PUMA_PICTURE_NONE 0
        # define PUMA_PICTURE_ALL 1

I tried all the table settings but didn't really get any different output. Setting it to anything other than 0 got it to OCR things inside a box outline (as opposed to turning the box into a picture).

[Jussi Pakkanen (jpakkane)]

This probably has to do with the table recognition code, which has not been open sourced yet. I'd like to get some comments from Cognitive people before committing this.

[Ben Jackson (ben.jackson)]

I saw the earlier discussion about this. I believe the missing code is for table *output* (eg as a spreadsheet). Enabling tables definitely allows the OCR code to look inside outlined areas, which it otherwise will not. In fact, if you OCR a page with a border it will not recognize anything without these new options.

Also, assuming bzr send is reasonable, the two patches are separate, so you could apply --nopictures.