As we are quite confident with the recognition rate, we’re doing some refactoring and optimization now. Because of the raw processing power modern GPU’s offer, we’re experimenting with HLSL Shaders for image processing.

The first tests showed a significant performance boost compared to similar things done on cpu.
Altough we’re just evaluating this method, we think that we can improve recognition performance dramatically.

Below a screenshot of a sobel filter implemented in a Pixel Shader applied to a 6 megapixel image: