Probabilistic matching and resemblance evaluation of shapes in trademark images

We present a novel matching and similarity evaluation method for planar geometric shapes represented by sets of polygonal curves. Given two shapes, the matching algorithm randomly generates a point sample from each shape and records a vote for a transformation which maps one sample to the other. The experiment is repeated many times. Clusters of votes in the transformation space indicate good candidate transformations for matching the two shapes. Unlike most voting schemes, though, the samples taken in one random experiment are extended as much as possible and the vote is weighted depending on the samples. The best clusters are those with a large total weight. The second part of the method is a resemblance evaluation of the two matched shapes. The definition of our resemblance function incorporates the proximity of line segments as well as the similarity of their slopes. The system is evaluated using the MPEG-7 shape silhouette database and a collection of 10 745 trade mark images. T...

• Type: Conference
• Year: 2007
• Where: CIVR
• Authors: Helmut Alt, Ludmila Scharf, Sven Scholz
• Keywords: CIVR 2007