Parallel implementation of automatic license-number extraction on a cluster of computers

The essential characteristics to achieve the reliable and useful rate of success in the automatic recognition of vehicle license-number using machine vision are the suitably controlled environmental setting and high recognition speed. Prior to the OCR operation, the location of license numberplate has to be found and the alphanumeric characters have to be successfully extracted. This paper presents another approach to develop a robust system to localize the license number area from a general image of a vehicle in a more natural way by means of variable thresholding based on fuzziness of the maximum entropy function. At each threshold, the potential character regions are tested according to the known properties of license numbers. The exhaustive searching is performed in parallel at each thresholding by various client-server processors connected in a high-speed local area network. The proposed method offers a straightforward approach for a more robust and flexible system to be developed.