Automatic license number extraction and its parallel implementation

This paper presents a segmentation technique based on multiple threshold values determined by fuzzy logic and information entropy principle to extract alphanumeric characters from general scene images, and its parallel implementation in a cluster of personal computers in a local area network. In this approach, the scene images are segmented into various regions based on fuzzy-entropy thresholding method which allows each local threshold to be detected and used for segmentation. A coarse searching technique is then implemented to locate potential character regions, followed by a rule-based character detecting technique to accurately extract characters. This searching step is executed in parallel at each computer for each selected threshold so as to collect all possible character-candidate regions. These results are then combined and x-y projection/profile is further applied to search for any missing characters. The multithreshold segmentation technique offers a simple, robust system that is capable of extracting alphanumeric characters from various scene images under nonuniform illuminating condition and for a variety of car number plates whereas the execution speed of the system is increased with parallel computing and dynamic load balancing technique.