2

I am looking for a method of cropping uploaded images. I spent a lot of time searching the site and find plenty of good solutions if the user is providing the dimensions/location of what is to be cropped.

My pictures are mostly slips of paper taken against a darker background. Think camera phone shots of drivers licenses, credit cards, ticket stubs against a somewhat darker (but not standard) background. I am looking for a method to either before or after upload to have the my code identify what should be cropped.

In a perfect world some smarts would make a suggestion as to what to crop and give the user a chance to give it a final okay. This would allow for a less then perfect cropping algorithm.

After looking up 'trim' instead of crop based on the first answer it seems like I could possible do this with Imagemagick imagemagick.org/script/api.php

Imagemagick seems to have intefaces in PHP & Ruby. It supports a trimming of 'fuzzy' images, one in which the boarder is not consistent. The problem that I am looking for advise on is a recommendation on a graphics engine that has good logic to find the edges of the images. Is there a better engine that I should be looking at then imagemagick?

Improve this question
2
  • If there is a way to have a uniform background of a color not appearing on the page, you could select background pixels fairly easily and use them to locate your crop boundary. Commented Feb 7, 2011 at 18:34
  • Your choice of "engine" depends on your target language. What are you using? Commented Feb 8, 2011 at 7:13

3 Answers 3

Reset to default
3

See if the following results are satisfactory. I implemented a small program in Mathematica to do it. If you like the results, I can explain further the algorithms, or you may check the Mathematica help files on line yourself.

We start with these images:

enter image description hereenter image description here

And define the following function: 

f[image_] := ImageCrop@ ImageMultiply[image, ColorNegate@ Erosion[Dilation[ DeleteSmallComponents[Binarize@ColorNegate@image, 10000], 3], 3]] // ImageAdjust

And the results are:

enter image description hereenter image description here

HTH!

Improve this answer
Sign up to request clarification or add additional context in comments.

2 Comments

Those results are pretty impressive. Would you mind explaining in a bit more detail? I would like to implement something similar--though probably using PIL instead of Mathematica.
@bjmc: Using only PIL will make the task hard for you, consider using at least scipy with opencv. All the steps in this Mathematica code are trivial, Binarize is using otsu thresholding, DeleteSmallComponents can be replicated using scipy.ndimage.label, erosion and dilation are readily available. But, this code fails for not so different images. Consider the image i.sstatic.net/SvoQe.jpg, the result with this exact code: i.imgur.com/xyVtj.jpg. Check stackoverflow.com/questions/14224995/….
1

Realize that, on a per pixel basis, the background you want to crop out is not consistent. Subtle (sometimes less than subtle) variations exist which you need to account for using a tolerance.

If you are familiar with Adobe Photoshop, there is a function that performs exactly what you are proposing (Image -> Trim). It works by looking at a user-selected corner pixel (top left, top right, etc.) and then determining the most outlying pixels that are not of that color.

In programming, you would need to iterate over the pixels in the image and determine the Y-coordinate of the topmost and bottommost pixel that is not of the selected color, the X-coordinate of the leftmost and rightmost. Those coordinates then determine the crop dimension.

When deciding if a pixel is or is not of the chosen color, you can introduce tolerance by allowing a percentage difference to factor out. In other words, if the background is not a consistent white (RGB 255, 255, 255), you may want to allow for some variation.

You didn't mention the programming language so it is up to you to research the graphic libraries and functions you may want to take advantage of.

Improve this answer

1 Comment

This was helpful... My first problem was I was thinking crop and not trim.. It seems like I could possible do this with imagemagick.org/script/api.php
0

One phrase which may be useful is "saliency" - identifying "interesting" areas of the image. And a place to start researching this is Microsoft Research - you can examples of images with the "interesting part" outlined. Does that achieve what you require?

Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.