||A practical vision system called VECON for VEhicle and CONtainer ID number recognition is described in this thesis. A robust technique for character extraction is proposed. The preprocessing locates potential character regions using selective line segment filtering allowing fast focus on real ID numbers. The potential characters regions are extracted through successive projection segmentation. Then the characters are extracted by a multi-extraction module integrated with variable threshold binarization. The extracted characters are finally grouped, normalized, and fed into a backpropagation neural recognizer for recognition. After being tested on more than 2000 container images and 2000 vehicle images under real scene conditions, and used probationally in an international container terminal, in several container depots and at the HKUST entrance, the system has proven to be fast, effective, and accurate. The system is PC-based and no dedicated hardware is needed.