Crop page from facing-page scan - opencv

Question

Suppose I want to crop just the left pages from facing-page scans of a spiral notebook like the example below (from Paolini.net).
Is there a more robust way than simply dividing the image's width by half? For example, a smarter algorithm would detect the spiral binding and make that the right boundary and even exclude black area to the left of the page.
If there's a relatively easy way to do this with OpenCV or ImageMagick, I'd love to learn it.

Answer 1

One possible way in ImageMagick 6 with Unix scripting is to do the following:
Trim the image to remove most of the black on the sides
Scale the image down to 1 row, then scale up to 50 rows just for visualization
Threshold the scaled image so that you get the black region down the spine as the largest black region
Do connected components process to find the x coordinate of the largest black region
Crop the image according to the results from the connected components
Input:
convert img.jpg -fuzz 25% -trim +repage img_trim.png
convert img_trim.png -scale x1! -scale x50! -threshold 80% img_trim_x1.png
centx=$(convert img_trim_x1.png -type bilevel \
-define connected-components:mean-color=true \
-define connected-components:verbose=true \
-connected-components 4 null: | \
grep "gray(0)" | head -n 1 | awk '{print $3}' | cut -d, -f1)
convert img_trim.png -crop ${centx}x+0+0 img_result.jpg
Data from connected components has the following header and structure:
Objects (id: bounding-box centroid area mean-color):
So head -n 1 gets the first black, i.e. gray(0) region which is the largest (sorted largest to smallest). The awk prints the 3rd entry, centroid, and the cut gets the x component.
If using ImageMagick 7, then change convert to magick
If you want to exclude the binders in the middle, then use the x-offset of the bounding box from the connected components listing:
convert img_trim.png -scale x1! -scale x50! -threshold 80% img_trim_x1.png
leftcenterx=$(convert img_trim_x1.png -type bilevel \
-define connected-components:mean-color=true \
-define connected-components:verbose=true \
-connected-components 4 null: | \
grep "gray(0)" | head -n 1 | awk '{print $2}' | cut -d+ -f2 | cut -d+ -f1)
convert img_trim.png -crop ${leftcenterx}x+0+0 img_result2.jpg
If you want just both pages, then we can find the white regions, i.e. gray(255) and crop them according to the width and x offset from the bounding boxes.
convert img.jpg -fuzz 25% -trim +repage img_trim.png
convert img_trim.png -scale x1! -scale x50! -threshold 80% img_trim_x1.png
OLDIFS=$IFS
IFS=$'\n'
bboxArr=(`convert img_trim_x1.png -type bilevel \
-define connected-components:mean-color=true \
-define connected-components:area-threshold=100 \
-define connected-components:verbose=true \
-connected-components 4 null: | \
grep "gray(255)" | awk '{print $2}'`)
IFS=$OLDIFS
num=${#bboxArr[*]}
for ((i=0; i<num; i++)); do
WW=`echo ${bboxArr[$i]} | cut -dx -f1`
Xoff=`echo ${bboxArr[$i]} | cut -d+ -f2`
convert img_trim.png -crop ${WW}x+${Xoff}+0 img_result3_$i.jpg
done