Legacy ImageMagick Discussions Archive

No, I meant the output should be the column (x of the image) that has the maximum value.

Scale the image to one row. Then output to txt: and sort by value. In unix, I would do it as

convert text_balloons_bw.png -scale x1! txt:- | tail -n +2 | sed 's/^\(.*\): .*gray[(]\(.*\)[)]$/\1 \2/g' | sort -r -k2

Code: Select all

or to get just the first one in the sorted list

Code: Select all

convert text_balloons_bw.png -scale x1! txt:- | tail -n +2 | sed 's/^\(.*\): .*gray[(]\(.*\)[)]$/\1 \2/g' | sort -r -k2 | head -n 1

577,0 99

Thanks Fred, that's what I wanted. I thought there might be a way for IM to spit it out, without using external tools, but that's fine.

This modification will eliminate the y=0 so only x and value:

Code: Select all

convert text_balloons_bw.png -scale x1! txt:- | tail -n +2 | sed 's/^\(.*\),.*: .*gray[(]\(.*\)[)]$/\1 \2/g' | sort -r -k2

Code: Select all

Actually the sed/sort snippet needs some fiddling because it lists two digits before three digits (99 before 113).

I also came up with these:

a. Is it possible to check for the color that is most scattered or "all over the place" (for a lack of a better word)? I want to identify the color of the background and then exclude it from the list of colors.

b. Or check the color that has the most continuous regions (i.e. the bubbles have the most continuous regions)? But not the most dominant color.

c. Maybe the background or the balloons can be found by taking note that the background takes the whole width, whereas the balloons take at most 80% of the width. Is there a way to check for these?

Edit: for clarity

This will fix the sort issue.

Code: Select all

convert text_balloons_bw.png -scale x1! txt:- | tail -n +2 | sed 's/^\(.*\),.*: .*gray[(]\(.*\)[)]$/\1 \2/g' | sort -n -r -k 2,2

Code: Select all

The only way to get continuous regions that I know is via connected-components. That is its purpose.

The only way I know to get the background color is too find the histogram color with the most counts, assuming the background color has a larger total area than any other color. So you can list the histogram and sort it by count.

P.S. Or just take the list of connected components and use the first entry, since it is sorted already be largest area first.

I edited my question after your post.

The sorting now works fine, thanks.

But connected components in only for thresholded images, right?

Your second post in this thread is what an implementation of the histogram color with the most counts right? The posterization doesn't work well with other images, e.g. a light gray that gets mixed with a white bg.

cso wrote: But connected components in only for thresholded images, right?

No, it works fine if you do not have too many colors or lots of fine detail such that you exceed 65536 regions on Q16 ImageMagick

cso wrote:Your second post in this thread is what an implementation of the histogram color with the most counts right? The posterization doesn't work well with other images, e.g. a light gray that gets mixed with a white bg.

No, my second post used your color image.

I did not posterize it. I simply reduced the number of colors using +dither -colors 8. I did that because you have a jpg image and its compression does not preserve large blocks of colors without potential changing the color values slightly. Therefore it might have exceeded the max number of regions of 65535 for Q16.

If I use 256 colors, then I get

Code: Select all

convert text_balloons.jpg +dither -colors 256 \
> -define connected-components:verbose=true \
> -define connected-components:area-threshold=5000 \
> -define connected-components:mean-color=true \
> -connected-components 8 \
> text_balloons_ccl2.png

convert: too many objects `text_balloons.jpg' @ error/vision.c/ConnectedComponentsImage/438.

Which shows that even for 256 colors with a jpeg, you get too many regions.

The same for 128 colors. 64 colors did not error, but was taking too long.

32 colors works OK

Code: Select all

convert text_balloons.jpg +dither -colors 32 \
> -define connected-components:verbose=true \
> -define connected-components:area-threshold=5000 \
> -define connected-components:mean-color=true \
> -connected-components 8 \
> text_balloons_ccl2.png
Objects (id: bounding-box centroid area mean-color):
  4720: 600x708+0+120 257.3,507.3 228315 srgb(252,252,252)
  4809: 412x204+173+133 376.3,236.0 79836 srgb(41,140,251)
  0: 520x120+0+0 222.8,49.8 47317 srgb(244,244,238)
  25915: 382x115+183+493 392.0,535.9 29530 srgb(41,140,251)
  37482: 301x99+78+637 232.1,686.2 27717 srgb(231,230,235)
  42455: 585x64+15+826 276.7,860.4 27319 srgb(252,252,252)
  26067: 415x122+170+506 353.6,593.9 21589 srgb(41,140,251)
  23299: 355x63+78+346 258.2,377.2 21067 srgb(231,230,235)
  42399: 600x74+0+826 336.4,866.2 16642 srgb(252,252,252)
  2604: 464x74+128+48 389.0,101.3 13367 srgb(235,246,251)
  1: 128x120+472+0 549.9,48.9 9977 srgb(252,252,252)
  41817: 160x64+422+746 500.7,778.1 8440 srgb(41,140,251)

But you have some other color regions showing as there are still small areas of white or black in the image inside the larger ones. So you probably should iterate a second time as I did in one of my other posts in your topic.

The main issue is how big to make the area threshold. That must be smaller than the size of the smallest area you want to preserve or find. That will be image dependent.

@cso: How are you capturing these images? They are graphic images, with areas of flat colour, but saved as jpg which uses compression designed for photographs, and changes colours. So the first task is to convert the image back to flat colours.

The job would be easier if the screengrabs were saved as png, which is lossless.

By posterization I actually meant color quantization.
Thanks for the explanation Fred, I will run some more tests and probably come back for more questions.
@snibgo: The actual images are png. I will experiment.

The images that one can download are JPG. So you should find a service that allows you to upload PNG without converting to JPG. If your images are PNG, you may still have many more colors than you want. You can check how many colors you actually have in your images using either the histogram or -unique-colors. See http://www.imagemagick.org/Usage/quantize/#extract. If you have few enough colors and not too many isolated regions of those colors, then you might not need to use the color quantization.

Legacy ImageMagick Discussions Archive

segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons

Re: segmentation of balloons