Legacy ImageMagick Discussions Archive

I wonder what would be the best way to use ImageMagick to identify a person (or a dog) walking over a grass in a sunny day, using a fixed camera.

For a cloudy day, that is not a problem, as you can see in the following image:

http://i1039.photobucket.com/albums/a47 ... vyhxjr.jpg

The image at the center is the reference image, while left image is the one with an intrusor (a man and a dog). In the right you can see the image obtained with the following command:

Code: Select all

    
    compare $f a0.jpg -fuzz 20% -metric FUZZ saida.jpg
    convert $f a0.jpg saida.jpg +append saida2_$f.jpg

$ f is the file that I want to examine.

As you can see, it worked fine, except for the situation when the grass at the background is partially iluminated by sun. See below:

http://s1039.photobucket.com/user/vpmam ... b.jpg.html

In the case above, the grass at the background is partially illuminated by sun, causing a very bright area which is identified as a feature by the difference algorithm.

I wonder if there is a way, in ImageMagick, to disregard shuch a effect using some sort of statistical technique. I would try some sort of co-variance method, but this is just a guess. Something like a variogram, or equivalent.

Is anybody interested on this topic?

Forgot to say that in the examples the dog and the man are inside the balcony. Just a detail.

Moreover, I forgot to say that a possible solution would be based on using a set of fixed reference images as the basis for comparison.

Here I am not concerned with camera movement, but background movement may be an issue... Right now I won't deal with background movement complexity and will consider that the background is fixed either.

Why not compare edge detected images? See
http://www.imagemagick.org/Usage/convolve/#sobel
viewtopic.php?f=4&t=25405

I will try. Thank you for the quick response.

Well, I tried your suggestion, but perhaps I used wrong technique.

See:

This image is devided in 3x2 areas: the upper line was obtained from the code below (already mentioned above):

Code: Select all

    
    compare $f a0.jpg -fuzz 20% -metric FUZZ saida.jpg
    convert $f a0.jpg saida.jpg +append saida2_$f.jpg

The bottom line was obtained by the code below:

Code: Select all

convert a0.jpg -define convolve:scale='50%!' -bias 50% -morphology Convolve Sobel:90 a0_e.jpg
convert a2.jpg -define convolve:scale='50%!' -bias 50% -morphology Convolve Sobel:90 a2_e.jpg
convert a2_e.jpg a0_e.jpg saida.jpg +append s_e.jpg

As you can see, at least through the method above, using edges is not meaningful.

Any other idea?

You need to threshold the edges or use the canny edge detector, which can remove short edges and less strong edges.

thank you!

would you be more specific?

how could I do this edge threshold?

I suggest the Canny edge detector, since it creates binary edges and can be tuned to remove short edges and weak edges. See viewtopic.php?f=4&t=25405

You could also post-filter your original compare results using -connected-components processing. See http://magick.imagemagick.org/script/co ... onents.php

ok, I will try.

thank you!

I will post the images here later.

EDITED: For connected-components processing you may want to use

Code: Select all

convert image1 image2 -compose difference -composite -threshold XX% result.gif

rather than compare. Then do the CCL.

Here is an example:

Code: Select all

convert image1.png image2.png -compose difference -composite diff.png

Code: Select all

convert diff.png -threshold 35% diff_t35.png

Code: Select all

convert diff_t35.png -define connected-components:verbose=true -connected-components 4 null:
Objects (id: bounding-box centroid area mean-color):
  0: 317x240+0+0 158.1,116.0 72563 gray(0)
  100: 92x68+109+172 156.2,207.7 2475 gray(255)
  101: 43x21+130+219 152.5,230.9 647 gray(0)
  35: 6x10+134+8 136.6,12.6 39 gray(255)
  5: 6x6+57+0 59.9,3.0 19 gray(255)
  7: 6x8+104+0 106.6,3.3 18 gray(255)
  6: 9x3+90+0 94.1,0.6 16 gray(255)
  11: 4x6+201+0 202.5,2.2 13 gray(255)
  9: 4x8+118+0 119.8,4.1 12 gray(255)
  104: 5x4+164+234 165.5,235.3 12 gray(255)
  28: 4x4+86+7 87.7,8.7 11 gray(255)
  57: 5x4+199+14 201.3,15.9 11 gray(255)
  19: 7x4+181+3 183.7,5.0 10 gray(255)
  3: 3x3+28+0 29.1,0.9 8 gray(255)
  20: 3x3+69+4 69.9,5.1 8 gray(255)
  1: 4x3+24+0 25.5,1.1 8 gray(255)
  27: 4x3+76+7 78.0,7.8 8 gray(255)
  44: 4x2+202+10 203.7,10.6 7 gray(255)
  16: 3x5+103+3 104.0,5.0 7 gray(255)
  33: 3x4+110+8 111.0,9.3 7 gray(255)
  62: 3x2+252+16 253.0,16.5 6 gray(255)
  42: 4x2+274+9 275.5,9.7 6 gray(255)
  55: 3x2+165+13 166.0,13.5 6 gray(255)
  82: 4x2+205+23 206.6,23.2 5 gray(255)
  105: 2x3+125+236 125.6,237.2 5 gray(0)
  91: 4x2+271+29 272.6,29.8 5 gray(255)
  79: 4x1+199+20 200.5,20.0 4 gray(255)
  75: 3x2+189+19 190.2,19.2 4 gray(255)
  86: 2x3+235+28 235.8,29.0 4 gray(255)
  77: 2x2+265+19 265.5,19.5 4 gray(255)
  60: 2x3+166+16 166.5,17.0 4 gray(255)
  80: 4x1+275+20 276.5,20.0 4 gray(255)
  84: 4x1+198+25 199.5,25.0 4 gray(255)
  74: 2x2+136+19 136.5,19.5 4 gray(255)
  39: 2x2+99+9 99.5,9.5 4 gray(255)
  76: 3x1+205+19 206.0,19.0 3 gray(255)
  87: 3x1+265+28 266.0,28.0 3 gray(255)
  81: 3x1+173+21 174.0,21.0 3 gray(255)
  17: 2x2+116+3 116.3,3.3 3 gray(255)
  95: 3x1+290+32 291.0,32.0 3 gray(255)
  65: 1x3+271+16 271.0,17.0 3 gray(255)
  98: 3x1+314+39 315.0,39.0 3 gray(255)
  78: 3x1+282+19 283.0,19.0 3 gray(255)
  51: 2x2+277+12 277.7,12.7 3 gray(255)
  64: 3x1+264+16 265.0,16.0 3 gray(255)
  31: 3x1+261+7 262.0,7.0 3 gray(255)
  45: 1x2+233+10 233.0,10.5 2 gray(255)
  43: 1x2+287+9 287.0,9.5 2 gray(255)
  48: 1x2+199+11 199.0,11.5 2 gray(255)
  73: 2x1+314+18 314.5,18.0 2 gray(255)
  32: 2x1+101+8 101.5,8.0 2 gray(255)
  41: 1x2+271+9 271.0,9.5 2 gray(255)
  2: 1x2+27+0 27.0,0.5 2 gray(0)
  70: 1x2+269+17 269.0,17.5 2 gray(255)
  10: 1x2+162+0 162.0,0.5 2 gray(255)
  88: 2x1+301+28 301.5,28.0 2 gray(255)
  34: 2x1+116+8 116.5,8.0 2 gray(255)
  66: 2x1+301+16 301.5,16.0 2 gray(255)
  58: 1x2+191+15 191.0,15.5 2 gray(255)
  12: 2x1+256+0 256.5,0.0 2 gray(255)
  103: 1x2+109+233 109.0,233.5 2 gray(255)
  102: 1x2+164+230 164.0,230.5 2 gray(255)
  46: 1x2+119+11 119.0,11.5 2 gray(255)
  107: 1x1+131+238 131.0,238.0 1 gray(255)
  4: 1x1+34+0 34.0,0.0 1 gray(255)
  8: 1x1+110+0 110.0,0.0 1 gray(255)
  13: 1x1+167+1 167.0,1.0 1 gray(255)
  14: 1x1+40+2 40.0,2.0 1 gray(255)
  15: 1x1+165+2 165.0,2.0 1 gray(255)
  18: 1x1+158+3 158.0,3.0 1 gray(255)
  21: 1x1+113+4 113.0,4.0 1 gray(255)
  22: 1x1+92+5 92.0,5.0 1 gray(255)
  23: 1x1+163+6 163.0,6.0 1 gray(255)
  24: 1x1+167+6 167.0,6.0 1 gray(255)
  25: 1x1+175+6 175.0,6.0 1 gray(255)
  26: 1x1+72+7 72.0,7.0 1 gray(255)
  29: 1x1+103+7 103.0,7.0 1 gray(255)
  30: 1x1+125+7 125.0,7.0 1 gray(255)
  36: 1x1+175+8 175.0,8.0 1 gray(255)
  37: 1x1+191+8 191.0,8.0 1 gray(255)
  38: 1x1+201+8 201.0,8.0 1 gray(255)
  40: 1x1+200+9 200.0,9.0 1 gray(255)
  47: 1x1+122+11 122.0,11.0 1 gray(255)
  49: 1x1+117+12 117.0,12.0 1 gray(255)
  50: 1x1+201+12 201.0,12.0 1 gray(255)
  52: 1x1+118+13 118.0,13.0 1 gray(255)
  53: 1x1+120+13 120.0,13.0 1 gray(255)
  54: 1x1+159+13 159.0,13.0 1 gray(255)
  56: 1x1+183+14 183.0,14.0 1 gray(255)
  59: 1x1+159+16 159.0,16.0 1 gray(255)
  61: 1x1+232+16 232.0,16.0 1 gray(255)
  63: 1x1+262+16 262.0,16.0 1 gray(255)
  67: 1x1+313+16 313.0,16.0 1 gray(255)
  68: 1x1+171+17 171.0,17.0 1 gray(255)
  69: 1x1+256+17 256.0,17.0 1 gray(255)
  71: 1x1+300+18 300.0,18.0 1 gray(255)
  72: 1x1+303+18 303.0,18.0 1 gray(255)
  83: 1x1+295+23 295.0,23.0 1 gray(255)
  85: 1x1+242+26 242.0,26.0 1 gray(255)
  89: 1x1+143+29 143.0,29.0 1 gray(255)
  90: 1x1+225+29 225.0,29.0 1 gray(255)
  92: 1x1+294+30 294.0,30.0 1 gray(255)
  93: 1x1+247+31 247.0,31.0 1 gray(255)
  94: 1x1+305+31 305.0,31.0 1 gray(255)
  96: 1x1+136+34 136.0,34.0 1 gray(255)
  97: 1x1+287+34 287.0,34.0 1 gray(255)
  99: 1x1+286+120 286.0,120.0 1 gray(255)
  106: 1x1+131+237 131.0,237.0 1 gray(0)

Now remove everything but the largest white/gray(255) area

Code: Select all

convert diff_t35.png -define connected-components:area-threshold=2475 -connected-components 4 -auto-level dog.gif

WOW!

Great!

I wonder if the threshold selected would work for all contrast conditions?

I will make a try with several images, different conditions, and will return here.

Thanks a lot! Hope this questions are useful for other people.

I am getting:

convert.im6: unrecognized option `-connected-components' @ error/convert.c/ConvertImageCommand/1107

Perhaps a problem with the IM version?

I tried the following code:

Code: Select all

#!/bin/bash
for f in D*.jpg
do 
    echo "Processing $f file.."
    convert $f a0.jpg -compose difference -composite saida.jpg
    convert saida.jpg -threshold 3% saida35.jpg
    convert saida35.jpg -define connected-components:area-threshold=2000 -connected-components 4 -auto-level saidafinal.jpg
    convert $f a0.jpg saida.jpg saida35.jpg saidafinal.jpg +append saida2_$f.jpg
    convert saida2_$f.jpg -append saida3.jpg
done;

Since the part related to "connected-components" is not working right now (see post above and the error message), I got only 4 files that I +appended to a row with 4 images.

See what I got with 3% of threshold.

What version of IM are you using? CCL was only added at 6.8.9.10

Legacy ImageMagick Discussions Archive

Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day

Re: Using ImageMagick to identify a person walking over grass for a fixed camera in a sunny day