Pattern recognition by distributed coding: test and analysis of the power space similarity method

This paper considers pattern recognition methods using distributed coding. These methods permit rapid learning from a large number of training samples; their recognition speed is high regardless of the size of the learning samples. This paper presents both basic algorithm and extended algorithms. Experiments with a large database of off-line handwritten numeric patterns are then described using the power space similarity method, being a type of distributed coding. Finally the effectiveness of the technique is considered.