We present here some results of our experimental evaluation.
For our first group of input images we used images from the MPEG-7 database with added noise. From all grouped contour segments, we selected the longest contour segment as the query. As can be seen, our method is able to exactly match the part segments with different invariance and occlusion, since all the top similar segments have the same class as the query segment. Although some other most similar segments are in different classes, they are visually similar to the query parts. All corresponding points found by shape context have similar neighborhoods, but the global appearance of matched segments is different in some cases. Shape context is based on similarity of histograms representing only local neighborhoods of the sample points, and it does not incorporate any shape representation related to a global appearance of compared objects.
To illustrate the object recognition potential on real images, we selected three color pictures. It is very interesting that our most similar result for the horse is an object in the MPEG-7 class called "carriage". This shows the limitations of keyword indexing of images, and it proves that our method can be useful for recognizing objects that, due to an overlap were merged with other objects, i.e., the horse and carriage, have one joint contour.
ResultsThe average time for part retrieval is approximately 2 seconds, using Matlab 6.5 on a 1.5 GHz Xeon-based PC. The time required to set up the part segment database and MSC database is about 30 minutes.
| <- Prev | Structural Shape Similarity -> |