image comparison (closed)

dinasset · **Posted:** Thu Mar 31, 2016 5:19 am **(#21)**

paynekj wrote:

So here's a suggestion for you to code up...

Both images as layers
Set the mode of the top layer to "Difference"
New layer from Visible (gimp-layer-new-from-visible)
With New layer active, look at the histogram (gimp-histogram)
Divide the Mean value by 2.55 to get a percentage (You may want to consider the means of all 4 channels)

Kevin

Kevin,
I'm on the way, but to better evaluate my results:
- could you post 2-3 (or just only 1 if you prefer) pairs of images indicating what is the percentage of similarity you get from ImageMagick? (to verify my tests)
thanks

paynekj · **Posted:** Thu Mar 31, 2016 7:14 am **(#22)**

dinasset wrote:

Kevin,
I'm on the way, but to better evaluate my results:
- could you post 2-3 (or just only 1 if you prefer) pairs of images indicating what is the percentage of similarity you get from ImageMagick? (to verify my tests)
thanks

Yes, I'll do that later today after I'm home from work. I'll also try and remember what the ImageMagick options are doing

Kevin

dinasset · **Posted:** Thu Mar 31, 2016 7:18 am **(#23)**

thanks, Kevin.

paynekj · **Posted:** Thu Mar 31, 2016 1:59 pm **(#24)**

Here are three files for comparison

And the results of the image magic command:

Code: Select all

Compare 2015-04-20-17-57-29.jpg to 2015-04-20-17-58-14.jpg difference=7.3
Compare 2015-04-20-17-58-14.jpg to 2015-04-20-18-00-02.jpg difference=2.7

If I do it manually using GIMP

Code: Select all

Compare 2015-04-20-17-57-29.jpg to 2015-04-20-17-58-14.jpg Value mean = 20.3 (/255 = 7.9%)
Compare 2015-04-20-17-58-14.jpg to 2015-04-20-18-00-02.jpg Value mean = 8.6 (/255 = 3.3%) 

And desaturating before comparing

Code: Select all

Compare 2015-04-20-17-57-29.jpg to 2015-04-20-17-58-14.jpg Value mean = 18.5 (/255 = 7.3%)
Compare 2015-04-20-17-58-14.jpg to 2015-04-20-18-00-02.jpg Value mean = 6.8 (/255 = 2.7%) 

Which matches the ImageMagick result.

Kevin

dinasset · **Posted:** Thu Mar 31, 2016 2:23 pm **(#25)**

Thanks Kevin.

These are my results (the output is the percentage of similarity:

1st vs 2nd : mean average similarity = 90.56
(average of the following 4 results from:
V,R,G,B histograms: 92.03 90.29 90.11 90.79

2nd vs 3rd : mean average similarity = 95.95
(average of the following 4 results from:
V,R,G,B histograms: 96.61 95.88 95.77 95.55

What is your opinion ?

paynekj · **Posted:** Thu Mar 31, 2016 2:36 pm **(#26)**

dinasset wrote:

Thanks Kevin.

These are my results (the output is the percentage of similarity:

1st vs 2nd : mean average similarity = 90.56
(average of the following 4 results from:
V,R,G,B histograms: 92.03 90.29 90.11 90.79

2nd vs 3rd : mean average similarity = 95.95
(average of the following 4 results from:
V,R,G,B histograms: 96.61 95.88 95.77 95.55

What is your opinion ?

I think it entirely depends on what you are going to do with the numbers.

For me I got enough difference to be able to automatically delete the 1500 images per day that didn't show anything different frame-to-frame.

Kevin

dinasset · **Posted:** Thu Mar 31, 2016 11:54 pm **(#27)**

OK.
It's enough for me, thanks again Kevin.