Pattern Recognition of Company Logos

Experimental Code from a small project that got interrupted due to lack of funding

In this prj we attempt to classify a number of Company Logos by assigning to each 2 distinct signatures extracted by a special horizontal and vertical scan of each image matrix. Each Logo picture is truncated to binary after resizing. This is done by the core routine imcount.m, while the main prog testpng.m runs a compelte test and displays results. Traces of each Logo could then be further processed by an appropriate learning system like a Neural Net, an Inference Engine, etc. The entropic-like index used for classification is better explained in the associated report downloadable from http://cag.dat.demokritos.gr/publications/TR2011-1.pdf.
The routine index0.m used for extracting the cluster analysis of each image matrix row/col can now be replaced by a most recent version "cluster.m", contained in another package "DEC2NARY" available by our Mathworks account.