| Summary: | digikam automatic photographic Labels | ||
|---|---|---|---|
| Product: | [Applications] digikam | Reporter: | Roger Larsson <roger.larsson> |
| Component: | Tags-Pick | Assignee: | Digikam Developers <digikam-bugs-null> |
| Status: | RESOLVED FIXED | ||
| Severity: | wishlist | CC: | caulier.gilles, julien.t43+kde, languitar |
| Priority: | NOR | ||
| Version: | 0.8.2 | ||
| Target Milestone: | --- | ||
| Platform: | unspecified | ||
| OS: | Linux | ||
| Latest Commit: | Version Fixed In: | 4.3.0 | |
| Sentry Crash Report: | |||
|
Description
Roger Larsson
2006-05-31 00:26:15 UTC
About sharpness: this can be problematic if only a small part of the image is sharp (like a macro shot of an insect) or a person before a blurred background. BTW, do you have a specific measure for sharpness in mind (i.e. how do you quantify: "only smooth edges")? There seem to be quite a few posssible ones around. I thinking about something like this: 1. Maybe convert to intensity greyscale image... 2a. Run a edge detector on the image 2b. Take absolute value of pixels 3. Make a histogram of the result 4. Compare histograms (top part) 5. Show result as a "analog" meter in tumbnail overview The highest scorer among similar pictures are probably the one with best focus - focus on the wrong part of a picture (like background) might result in erronous higher scores. But those can probably easily be found when viewed as thumbnail. 6. Improve algorithm... (eye/face recognition)... Roger,
I think this is very interesting, and running a search
for "detect blur" gives many hits.
For example there is a paper "Blur Detection for Digital Images Using Wavelet Transform" which sounds interesting as it will output a single
number ("blur extent"). However, I am not sure whether the given examples
really correspond to "real world" situations.
So, before anything interesting like this is integrated into digikam,
some research is needed, in my opinion.
We would need a stand-alone implementation for one (better: a bunch) of
the methods (including the one you proposed, of course) and
see how well they perform on real world images.
To test such code, it might be better to prototype
it eg. in python. Once could think of using the
Python Image Library (http://www.pythonware.com/products/pil/)
(together with numpy) or VIPS (http://www.vips.ecs.soton.ac.uk/index.php?title=VIPS)
which is also scriptable using python.
Of course one can do the stuff straight in C++ ... ;-)
Do you think you could maybe have a try with that?
(For the python/PIL/numpy I could give you some more pointers, if needed).
Another automatic tag I need: portrait/landscape This time every year I go through this years photos to find twelve to put on a calendar for my mother. But they really have to be in landscape format or they won't look good... This could be solved with a Batch Queue Manager tool? yes, and camera gui, during items download. Warnings : BQM using multithreading, and this feature need to acesss to database. SQlite is not really Tread-safe. Take a care... Note : This problem will be solved when Mysql support will be completed... Gilles Speaking in general: Since we are at a "final" beta state, we shouldn't implement new features anymore. I already think digiKam is taking "beta" not as serious as other applications :-) Beta means: Here is the "almost" final stage, what do you think? But we have like 6 beta versions now and continue to add new stuff. We can not handle all the new bugreports for the short RC stage then. I didn't plan to introduce this now. ;) It was just a general thought about that topic. from http://mail.kde.org/pipermail/digikam-users/2012-June/016417.html other usages: source pictures to create panoramic or HDR composition. two needed things: * search pictures based on complex conditions (group of meta / actions / calculus or external program results). For me, in this kind of case, a free unique search field (a la google) with boolean and parenthesis would be more efficient than gui for each field. Also one of the could-be problem is be able to make selection base on image meta and "nearby" images (be it near in time, near in sharpness or whatever) * 'autotagged' them either after some actions (like import, export or edit), either on demand This feature have been already implemented in 4.x release through Image Quality Shorter tool. See config panel dedicated for details. Note this tool still in experimental state. Please test and report. Gilles Caulier |