I have a set of images (e.g. cars) and trying to perform feature detection using SIFT (actually, SIFT modification - A-SIFT, which is more robust to affine transformations) and I'm getting an extremely low results in matching even on the same models, same color and closely same view point.
I'm barely sure, that the problem is car's surface is glossy (that's leads to huge variation of reflections on surface) and shadows & lighting (that's depends on location where photo was taken). Is there's image processing algorithms that can help me clean up images and/or decrease influence of that factors?
I had a quick look at CiteSeerX and Google Scholar but found nothing.
$\begingroup$
$\endgroup$
8
-
1$\begingroup$ The question is probably more suitable for MetaOptimize or Stats.SE. $\endgroup$– KavehJul 16, 2011 at 14:41
-
$\begingroup$ I'm not quite sure about MetaOptimize and Stats - question isn't about noticeably large amounts of data and complicated decisions inherent to machine learning. At least not directly. Nor it refer to theoretical computer science, but I've seen here a couple computer vision questions so I've placed it here. $\endgroup$– om-nom-nomJul 16, 2011 at 16:31
-
3$\begingroup$ I'm afraid this isn't quite the right place for this question (which I personally find somewhat interesting). $\endgroup$– Suresh VenkatJul 16, 2011 at 17:56
-
$\begingroup$ IIRC, CV is in the scope of MetaOptimize. $\endgroup$– KavehJul 16, 2011 at 22:34
-
$\begingroup$ @kaveh, @suresh-venkat, I'll try to ask my question at more general thematic Stack Overflow. $\endgroup$– om-nom-nomJul 17, 2011 at 1:06
|
Show 3 more comments
1 Answer
$\begingroup$
$\endgroup$
A common preprocessing tactic is to smooth the image prior to feature extraction using for instance a Gaussian kernel or a low pass filter. This allows for the high frequency spurious details to be removed.
