Finding vertical stipes to use for crop input

[TedBaker]

I have images of films strips from a scanner, like the following example. These, that have been nicely trimmed.

https://ibb.co/dWDTbR

I want to find the horizontal coordinates of the separating strips, i.e. the continues grey areas separating each frame. I want to use the horizontal coordinates to drive a cropping operation. That way I can carefully control the cropping with my own logic.

I have a reasonable estimate of the 'colour' of each vertical spacing gap, in this case gray(35%). So I am looking for the horizontal coordinates of continuous area that matches the said colour with fuzzy match of 5% or so. Ideally of would like to know the start and end range of those coordinates.

Any clues?

Any help appreciated.

IM 6.8.9 linux

[fmw42]

Is this not just a repost of viewtopic.php?f=1&t=32821

[TedBaker]

Is this not just a repost of viewtopic.php?f=1&t=32821

No the problem is more specific, hence I thought it best to start a new topic.

[snibgo]

You can follow the approach Fred gave in today's thread viewtopic.php?f=22&t=32928, but horizontally instead of vertically. That is, scale to a single row, and auto-level. If you hadn't whited out portions, the margins would be the lightest portions of the image.

[fmw42]

The problem is that you have shades of gray elsewhere of the same value. I can get it close, but there may be false positives that you will have to filter by looking at the regularity of the gaps.

The process is to

Replace near white with black to remove the white blobs so they do not average with the rest to similar shades of gray
Average the image to one row and then expand it again.
Replace your gap color with white and everything else with black
Use morphology to remove stripes of 1 pixel width as a precaution
Use connected components to the find the bounding boxes of all separate white regions.

Code: Select all

dim=`convert test.jpg -format "%wx%h" info:`
convert test.jpg -fuzz 25% -fill black -opaque white tmp0.png
convert tmp0.png -scale x1! -scale $dim! tmp1.png
convert tmp1.png -fuzz 2% -fill white -opaque "gray(90)" -fill black +opaque white tmp2.png
convert tmp2.png -morphology open "5x1:0,1,1,1,0" tmp3.png
arr=()
OLD_IFS=$IFS
IFS=$'\n'
arr=(`convert tmp3.png -type bilevel \
-define connected-components:verbose=true \
-define connected-components:mean-color=true \
-connected-components 4 \
null: | sed 's/^[ ]*//' | tail -n +2`)
IFS=$OLD_IFS
num=${#arr[*]}
echo "num=$num"
for ((i=0;i<num;i++)); do
bbox=`echo "${arr[$i]}" | cut -d\  -f2`
color=`echo "${arr[$i]}" | cut -d\  -f5`
if [ "$color" = "gray(255)" ]; then
echo "bbox=$bbox;"
fi
done

original:


tmp0 -- remove white blobs


tmp1 -- average to one row and expand


tmp2 -- convert to black and white


tmp2 -- use morphology to remove 1 pixel wide stripes (not needed this time)


result:

Code: Select all

bbox=10x94+716+0;
bbox=10x94+716+0;
bbox=8x94+868+0;
bbox=7x94+277+0;
bbox=6x94+130+0;
bbox=5x94+570+0;
bbox=5x94+424+0;

[TedBaker]

I can get it close, but there may be false positives that you will have to filter by looking at the regularity of the gaps.

Ok that's absolutely amazing, or should I say absolutely magic! The false positives should not a problem, to filter out as I can filter those out against the known aspect ratio of the frames, which is why I wanted to do it this way.

Thanks again!