MobileRead Forums - View Single Post

ereszet · 10-01-2007, 09:01 AM

Cacapee, below is a list of refinements for you to consider to make your program a comprehensive tool for "paper books to image" processing:

1. Deskew (based on lines rather than borders - like ClearImage)
2. Clean Black Borders before wiping out White Borders (again like ClearImage that removes the borders, and sets border area pixels to white). That option is reliable for B&W pictures only.
3. Straighten the curved text lines (the only program I know that does it properly is BookRestorer)
4. Remove the perspective/trapezoidal effect (as above)
5. Despeckle/ Remove noise
5. Remove black objects (lines and blobs) with a specified number and density of black pixels. No program that I know can do it properly.Some programs allow to remove the punch holes (ClearImage) and some allow to remove lines or blobs of certains dimensions (Scanfix) but every so often they remove parts of text as well. That is why I suggest to take into account the number of black pixels that are close together rather than vertical and horiontal dimensions of black objects. The line of text is a mix of black and white pixels, while a black blob will have much higher proportion of black to white.

I do not expect you to do what professional/commercial teams of developers have not achieved yet, but it looks from what you have already done that you can improve on them. The cleaning of black borders would be the most welcome refinement for "picture perfect" photos of book pages (camera pictures are of different proportions than paper documents/books and that leaves the black bacground at the picture borders). The other refinements are for poorly taken photos/microfilms from digital libraties.

Once again let me tell you how much I admire what you have done, and my suggestions are just a dream for the future.

10-01-2007, 09:01 AM	#218
ereszet Zealot Posts: 118 Karma: 306 Join Date: Sep 2007 Device: Sony PRS-500 Archos 704 wifi	Refinements to consider - paper book to image Cacapee, below is a list of refinements for you to consider to make your program a comprehensive tool for "paper books to image" processing: 1. Deskew (based on lines rather than borders - like ClearImage) 2. Clean Black Borders before wiping out White Borders (again like ClearImage that removes the borders, and sets border area pixels to white). That option is reliable for B&W pictures only. 3. Straighten the curved text lines (the only program I know that does it properly is BookRestorer) 4. Remove the perspective/trapezoidal effect (as above) 5. Despeckle/ Remove noise 5. Remove black objects (lines and blobs) with a specified number and density of black pixels. No program that I know can do it properly.Some programs allow to remove the punch holes (ClearImage) and some allow to remove lines or blobs of certains dimensions (Scanfix) but every so often they remove parts of text as well. That is why I suggest to take into account the number of black pixels that are close together rather than vertical and horiontal dimensions of black objects. The line of text is a mix of black and white pixels, while a black blob will have much higher proportion of black to white. I do not expect you to do what professional/commercial teams of developers have not achieved yet, but it looks from what you have already done that you can improve on them. The cleaning of black borders would be the most welcome refinement for "picture perfect" photos of book pages (camera pictures are of different proportions than paper documents/books and that leaves the black bacground at the picture borders). The other refinements are for poorly taken photos/microfilms from digital libraties. Once again let me tell you how much I admire what you have done, and my suggestions are just a dream for the future.