Research On License Plate Tilt Correction And Character Recognition

License plate automatic recognition system is divided into vehicle license plate location, license plate tilt correction, license plate character segmentation and license plate character recognition. In this system, those four key parts directly affects the recognition rate of license plate automatic recognition system. Although, more and more attention at automatic license plate recognition technology has brought many pioneering achievements, but also higher demand appeared, especially the automatic license plate recognition of freedom in complex environment, urgently need better algorithms to solve more complex problems. Therefore, automatic license plate recognition technology has broad prospects for development.This paper introduces the key technology of automatic license plate recognition according to the general process of automatic license plate recognition. First of all, the license plate slant correction and the license plate character recognition algorithms are studied in theory. The two algorithms, the license plate tilt correction based on rotating projection pose evaluation function and the recognition method for license plate characters based on scoring model, are proposed. And validate the algorithms through the experiments. In this paper, the main research contents as follows:(1) The license plate tilt correction based on rotating projection pose evaluation function is proposed. The methods of common license plate tilt correction have some advantages and disadvantages, generally the complexity of the algorithm is high, so as to the operation time, and when license plate character similar to the color of the non plate region color or there are some stains on the license plate, the license plate tilt correction accuracy is not very ideal. In order to overcome these disadvantages, firstly license plate images turn to binaryzation. Then the image of binaryzation is converted to jump point diagram, that is to scan the license plate images line by line, when the black or white black jump at this point, the pixel value is set to "1", the other pixels set to "0". Then the jump point diagram is projected on the direction of perpendicular, found that when the tilt angles change, the projection posture will appear some changes, according to these changes to construct rotating projection pose evaluation function, then uses the evaluation function to correct the tilt of the license plate, and finally through the experiment to verify the effectiveness of the algorithm.(2) The recognition method for license plate characters based on scoring model is proposed. The license plate character image that cutting from the surveillance video tend to have some deficiency, for example low pixel, fuzzy, character confusion serious or edge uneven. The traditional license plate character recognition algorithm for this type of image has low recognition rate, in order to improve that recognition rate. The character image is divided into some grids firstly in the algorithm, and then the grids are divided into different grades according to the number of the white point in the grid, so is the standard character template. When identify matching, the corresponding square of the recognition character picture and the character template are grade matching according to the grade difference, and then score in accordance with the scoring model and the graded, plus, minus or zero. Finally add up the scores, the character template of high scores will be the final recognition results. The experimental results verify the effectiveness of the algorithm.