Hello,

I would like to obtain a black and white image of a photograph of a book page.

My attempt to do this is by running the threshold function over the entire image. This appears to give the desired result, but only locally. It appears that the optimum parameters of the threshold function vary by position (i.e., according to some scalar field), with one extreme in the center of the glare and then radiating outward. So just running the threshold function over the entire image with constant parameters always gives sub-optimum contrast for 75% of the image. Will I have to find a way to run the threshold function with non-constant parameters, or is there some other tool to do the job?

Thanks for reading, Nick

Hello,

I would like to obtain a black and white image of a photograph of a book page.

My attempt to do this is by running the threshold function over the entire image. This appears to give the desired result, but only locally. It appears that the optimum parameters of the threshold function vary by position (i.e., according to some scalar field), with one extreme in the center of the glare and then radiating outward. So just running the threshold function over the entire image with constant parameters always gives sub-optimum contrast for 75% of the image. Will I have to find a way to run the threshold function with non-constant parameters, or is there some other tool to do the job?

Thanks for reading, Nick

Thanks for the replies. I've uploaded the original image, along with three modified images that are satisfactory at various distances along my imaginary gradient.

original: http://lh5.googleusercontent.com/-zw3ICTEiurs/T3fwShHiK8I/AAAAAAAAAKU/fTTWnYCJyZA/s575/original.jpg

contrast applied with increasing upper levels parameter and minimal gamma (no gray)::
http://lh6.googleusercontent.com/-Vc0C2KcpOAw/T3fwSLBrexI/AAAAAAAAAK4/9AEbK7qyBHc/s771/1.JPG http://lh5.googleusercontent.com/-6x8X75cw4JI/T3fwSmRcODI/AAAAAAAAAKY/LfRwru-LQQg/s771/2.JPG http://lh4.googleusercontent.com/-48qQsSkULZU/T3fwSJMN5pI/AAAAAAAAAKI/r5AdxIcFvb0/s771/3.jpg

You can see from the trend that there does exist a set of parameters for the threshold function that gives acceptable contrast for every given point, and that this one parameter (the upper level) increases monotonically as it moves along the "vector field".

Regarding the suggestions: I think I've played with levels enough without results. On the other hand, I don't know about the layer mask approach. From my limited abilities, I can only see that layer masks modulate the opacity of the resulting image after having applied some filter to a duplicate layer, whereas I want to modulate the input parameters of a "filter" (threshold) to the original layer.

Sorry if I'm making this more complicated than necessary, and thanks for reading.

-Nick

On 3/31/12, Owen wrote:

Hello,

I would like to obtain a black and white image of a photograph of a book page.

My attempt to do this is by running the threshold function over the entire image. This appears to give the desired result, but only locally. It appears that the optimum parameters of the threshold function vary by position (i.e., according to some scalar field), with one extreme in the center of the glare and then radiating outward. So just running the threshold function over the entire image with constant parameters always gives sub-optimum contrast for 75% of the image. Will I have to find a way to run the threshold function with non-constant parameters, or is there some other tool to do the job?

Without seeing the image, one can only suggest you try converting the image to Indexed and 2 colors.

You could also leave it as rgb and play with the curves, maybe one of the colors can be filtered out.

If you get a good B & W image, you might then want to run it through the unsharp mask to improve the text.

-- Owen

Here is the way I compensate for such exposure when you know the background should all be one colour:

1) Duplicate your source layer. 2) Select the duplicate layer and run Filters->Generic-> Dilate. 3) Repeat the dilate (ctrl-f) until the (black) print is gone. (For your sample I did it three times)
4) Apply a larger radius Gaussian blur (10 times the number of dilates you did seems to work for me, so in this case, 30px) 5) This layer is kind of a low-pass filter version of your page with the text removed. Set this layer to grain extract to kind of remove it from the original.
6) Do Layer->New from Visible to get the composite. 7) Now go and play with levels/curves/thresholding.

Note this works best with low noise images (i.e. tiff or png is better than jpeg) and higher resolution scans...

Here is what I ended up with: http://www.majhost.com/gallery/ffaat/gimp/more2/results.jpg

-Rob A>

Thanks, this method gives virtually perfect results. Also, thanks to everyone for writing your suggestions.

I like how the generality of the "grain extract" and its siblings would seem to make it a powerful tool in a number of areas, and I'm now fascinated. :-) I wonder if there exists a sort of compendium of gimp techniques from the signal processing perspective....

Nick