| Summary: | 碩士 === 國立臺灣科技大學 === 機械工程系 === 99 === Vehicle license plate recognition is generally composed of three modules: license plate detection, character segmentation and character recognition. This research focuses on the improvements of character segmentation and recognition modules. The characters are often wrongly segmented because of shadows, unbalanced or extreme illumination on the plates, or large viewing distances or angles from the cameras. This research applies MSER (Maximally Stable Extremal Region) detector to extract the interest regions among the characters, which are then processed by connected components to determine the best boundary to segment the characters. Given accurately segmented characters, the characters with similar global features are relatively easy to be misclassified, for example, “O” and “D”, and “8” and “B”. A hierarchical classifier is proposed in this research which is composed of a main layer and an enhancement layer. The main layer extracts the global features from all of the alphanumeric characters, and can classify those with distinctive global characteristics. The enhancement layer is only meant for the groups of characters with similar global features by using the group-specific features for more refined classification. The proposed methods are evaluated on the AOLP (Application Oriented License Plate) benchmark database together with a comparison to other competitive methods. Experiments show that the proposed methods can effectively improve the performance of both the character segmentation and recognition modules.
|
|---|
