SportWagon
March 2nd, 2015, 01:06 PM
It had previously seemed to me like GraphicsMagick (https://en.wikipedia.org/wiki/GraphicsMagick) was a more freely licensed version of ImageMagick (https://en.wikipedia.org/wiki/ImageMagick).
Mumble, but, consulting wikipedia, it seems like both are pretty freely licensed now. As if sometime after I'd been tangentially involved in carefully installing GraphicsMagick instead of ImageMagick in environments to avoid licensing problems, the concern has gone away.
Bah. Of course, issues still do arise when the programs do or don't manipulate graphics formats which themselves are not free.
Anyway, I'd had a reason to see if simple command-line image recognition
If the source of your images is in-game screenshots, such simple analysis does seem to work relatively well for identifying "the type of screen depicted". So after having gathered lots of shots without classifying them I should be able to relatively easy pick shots of a type I want by using "compare" to evaluate them to a number indicative of their "closeness".
Using "compare" (imagemagick) or "gm compare" (graphicsmagick) I seem to get results corresponding to what I expect. Or at least, after a few samples I can deduce what I should use as a criterion for a "match".
Reading the wikipedia pages seems to suggest I would be better to go with GraphicsMagick (https://en.wikipedia.org/wiki/GraphicsMagick) rather than ImageMagick (https://en.wikipedia.org/wiki/ImageMagick). Because GraphicsMagick seems to me more concerned about a useful command-line/API. Bah. But the most natural shared environment for me to do my processing on seems to have ImageMagick but not GraphicsMagick.
It seems to me going with either "MAE" (Mean Absolute Error) or "RMSE" (Root Mean Squared Error) evaluation (seem to be supported by both packages) give recognizable results, in small test cases. But RMS is so traditional.
Anybody have other comments and suggestions.
This is purely a personal project, nothing to do with work.
Mumble, but, consulting wikipedia, it seems like both are pretty freely licensed now. As if sometime after I'd been tangentially involved in carefully installing GraphicsMagick instead of ImageMagick in environments to avoid licensing problems, the concern has gone away.
Bah. Of course, issues still do arise when the programs do or don't manipulate graphics formats which themselves are not free.
Anyway, I'd had a reason to see if simple command-line image recognition
If the source of your images is in-game screenshots, such simple analysis does seem to work relatively well for identifying "the type of screen depicted". So after having gathered lots of shots without classifying them I should be able to relatively easy pick shots of a type I want by using "compare" to evaluate them to a number indicative of their "closeness".
Using "compare" (imagemagick) or "gm compare" (graphicsmagick) I seem to get results corresponding to what I expect. Or at least, after a few samples I can deduce what I should use as a criterion for a "match".
Reading the wikipedia pages seems to suggest I would be better to go with GraphicsMagick (https://en.wikipedia.org/wiki/GraphicsMagick) rather than ImageMagick (https://en.wikipedia.org/wiki/ImageMagick). Because GraphicsMagick seems to me more concerned about a useful command-line/API. Bah. But the most natural shared environment for me to do my processing on seems to have ImageMagick but not GraphicsMagick.
It seems to me going with either "MAE" (Mean Absolute Error) or "RMSE" (Root Mean Squared Error) evaluation (seem to be supported by both packages) give recognizable results, in small test cases. But RMS is so traditional.
Anybody have other comments and suggestions.
This is purely a personal project, nothing to do with work.