New Approach To Recognition Of Blurred License Plate Based On Improved Natural Image Matting Algorithm

Car's license plate identification system based on image processing is one of the key technologies of intelligent transportation system, and it plays an important role in transporting system. Though recognition rate of license plate are above 90%, as a result of the complex environment, when the license plate are seriously blurred by some dunghill such as mud or water smoke before camera lens, it is hard to be identified. In order to get higher recognition rate of blurred license plate, it is necessary to propose these blurred license plate.License plate location, character segmentation and character recognition are included during the recognition steps. The research in this paper consists of license plate location, character segmentation and character recognition. Edge detection with robbert operator and mathematical morphology are combined in location step, which is fit for simple scene; Then character divisions are finished by character projection segment; In recognition phase, simple template matching is adopted here, which finds the most similar temple to current character. In most cases, clear license plate numbers can be accurately recognized by the generic identification systems. However, sometime the blurred license plate is seriously hard to be identified, which are necessary to be proposed before being recognized.In order to recognize blurred license plate, an approach based on improved natural image matting is proposed in this paper. After those fuzzy license plates are processed by natural image matting, clearer plate numbers are obtained so that it can been identified accurately. Then they are recognized by simple template matching. At first, it is necessary to provide the information about foreground and background by paintbrush with a sparse brush image. At most cases, extraction is finished when solve the F andαassociated with I. And then, we must figure out the background B. Lastly, in order to get better result, normalize B's brightness to I's .Because it has been darkened.Finally, for blurred image covered by foreground, as the experiment is shown that by means of natural image matting, performance of blurred license plate recognition can be obviously improved.