RMagick image conversion

Question

While hacking a simple project I'm stuck with a RMagick problem for a day now. Since I am very new to RMagick need suggestions on this.

I am trying to scan a user uploaded image and try to snap/select the image to specific region where actual content is. But there is a chance that user can take image with phone or other device and end up adding much noise to image.

Example image uploaded from user looks like this.

I tried to convert the image to a greyscale like bellow

 img.quantize(2, GRAYColorspace, false)

But which has so much of noise when converted (image). Could any one suggest how can I select only the section/rectangle which contains actual information from the uploaded image?

Bellow are the methods I'm trying to get my hands dirt with.

• get_pixels
• find_similar_region
• adaptive_threshold

Answer 1

In ImageMagick command line, you can use -negate (invert colors) and -lat (local area threshold) to convert this to binary (black/white), which might be more what you want. There should be equivalent commands in RMagick.

Input:

convert t2.jpg -negate -lat 20x20+10% -negate result.png

ADDITION:

Note, there are a few dark spots in the resulting image above. You need to filter those using -connected-components before you can trim the image to the black box.

convert t2.jpg -negate -lat 20x20+10% -negate -type bilevel \
-define connected-components:area-threshold=10 \
-define connected-components:mean-color=true \
-connected-components 4 \
-bordercolor white -border 1x1 \
-fuzz 15% -trim +repage \
result.png