How to extract photos from JPG with white background? - image

Question

I have a JPG file which contains multiple photos over a white background.
I'm looking for a CLI tool which can extract photos from the source JPG (without supplying coordinates) into separate JPG files maintaining quality and photo resolution.
From some research I suspect ImageMagick can achieve this though unsure of the correct CLI command.
If useful, I'm on OSX 10.13.2 and have ImageMagick 7.0.7-28 installed.

Answer 1

Here are two ways to do this in Unix using Imagemagick. I just cropped out your base image from your diagram, since I was not sure it that was part of your image or not. If it is part of the image, then you would have to trim that off first using -trim.
Input:
The first is my script, multicrop2:
(-f 10 is the fuzz factor for extracting the background)
(-u 3 means no attempt to unrotated the results)
multicrop2 -f 10 -u 3 image.jpg resulta.jpg
Processing Image 0
Initial Crop Box: 113x84+81+89
Processing Image 1
Initial Crop Box: 113x67+144+10
Processing Image 2
Initial Crop Box: 113x66+10+11
The second is using Imagemagick -connected-componets (which is what I use in my script)
What this does is:
1) fuzzy flood fill the background to transparent (since jpg is loss and does not preserve a uniform background.
2) change the color under the transparent to white and remove the transparency
3) change anything not white to black
4) apply -connected-components to throw out areas smaller than 400 pixel area and extract each bounding box and color
5) if the color is gray(0), i.e. black, then crop the original image to the bounding box and save to disk
OLDIFS=$IFS
IFS=$'\n'
arr=(`convert image.jpg -fuzz 10% -fill none -draw "matte 0,0 floodfill" \
-background white -alpha background -alpha off \
-fill black +opaque white -type bilevel \
-define connected-components:verbose=true \
-define connected-components:mean-color=true \
-define connected-components:area-threshold=400 \
-connected-components 4 null: | tail -n +2 | sed 's/^[ ]*//'`)
IFS=$OLDIFS
num=${#arr[*]}
j=0
for ((i=0; i<num; i++)); do
bbox=`echo "${arr[$i]}" | cut -d\ -f2`
color=`echo "${arr[$i]}" | cut -d\ -f5`
if [ "$color" = "gray(0)" ]; then
convert image.jpg -crop $bbox +repage resultb_$j.jpg
j=$((j+1))
fi
done
EDIT: Add processing of your actual image
Input:
The first thing to note is that your actual two images are right at the right side, but there is a black edge there. Also one at the top. That black edge connects the two image so that they cannot easily be separated by the multicrop2 script. So you need to shave off the right side by enough pixels to remove that edge. There is also the edge at the top and you can shave that off if you want. If you do, you can reduce the -d argument. The -d argument needs to be smaller that the area of the smallest image you want to extract and larger than any other minor noise or the stripe at the top in area. So I clip off 20 pixels from the right side and then use multicrop2 with a very large value for -d. I have chosen a value for -f of 8, which is seems to be in a rather narrow range due to the non-constant background. You can add -m save to look at the mask the script creates to see that you get a good separation between your two images. I seed the processing at -c 20,20 to avoid the black border at the top of your image, so that the script gets a good measure of the background color for the flood fill step.
convert test.jpeg -gravity east -chop 20x0 tmp.png
multicrop2 -c 20,20 -f 8 -d 100000 tmp.png result.jpg
Processing Image 0
Initial Crop Box: 2319x1627+968+2153
Processing Image 1
Initial Crop Box: 2293x1611+994+436