Quantization of the Space of Structural Image Features as a Way to Increase Recognition Performance

摘要

A modification of the structural image recognition method in computer vision systems is proposed. In order to improve the performance of recognition, quantization (clustering) is applied in the space of characteristic features that form the pattern of the object. Due to the transformation of structural objects descriptions from a set representation to a vector form, the amount of computation is reduced tens of times. The results of experiments that confirmed the effectiveness and increase of decision-making process are shown.