CN103390160A

CN103390160A - License plate character segmentation algorithm based on connected domain and template

Info

Publication number: CN103390160A
Application number: CN201310317924XA
Authority: CN
Inventors: 吴志伟; 冯琰一; 张少文
Original assignee: PCI Suntek Technology Co Ltd
Current assignee: PCI Suntek Technology Co Ltd
Priority date: 2013-07-24
Filing date: 2013-07-24
Publication date: 2013-11-13

Abstract

The invention discloses a license plate character segmentation algorithm based on a connected domain and a template. The method comprises the following steps: firstly performing graying on corrected license plate images, synthetically utilizing an OTSU automatic threshold value method, a block region growth method and a gradual threshold value improving method to extract a prospect of characters, and using the connected domain and prior knowledge of the characters to generate a superior segmentation character library; performing averaging on the width and the height of all characters in the superior segmentation character library to generate the standard license plate template, synthetically utilizing large spacing and information to perform character segmentation, using gradient sums to evaluate all segmentation results, and choosing a segmentation result with the maximal gradient sum as the optimal segmentation result. The character segmentation algorithm disclosed by the invention can well process situations of uneven illumination, multimodal characteristics shown in inter-class variance distribution, mixing of front character background colors and the like.

Description

A kind of cutting of characters on license plate based on connected domain and template algorithm

Technical field

The present invention relates to computer vision field, relate in particular to a kind of method of the cutting of the characters on license plate based on connected domain and template.

Background technology

Along with the raising of Recognition of License Plate Characters accuracy rate, character cutting has become the key issue in Vehicle License Plate Recognition System.The accuracy rate of character cutting and the accuracy rate of character recognition have direct relation, and the mistake of character cutting tends to cause the mistake of character recognition.Accuracy in the situations such as accuracy rate, the especially uneven illumination that therefore how to improve character cutting is even, characters on license plate is unintelligible, become the key of the accuracy rate of the whole Vehicle License Plate Recognition System of further raising.

Present stage, the most frequently used characters on license plate cutting method was vertical projection method.The method utilizes the gap of character vertical projection to carry out the cutting of character, and English License Plate Character Segmentation effect is better preferably for illumination condition, but to the easy cutting mistake of characters on license plate in the situation such as Chinese character or Characters Stuck, uneven illumination be even.

Summary of the invention

The invention provides a kind of new characters on license plate cutting method, improved the accuracy of character cutting.

The present invention adopts following technical scheme:

A kind of cutting method of characters on license plate based on connected domain and template comprises:

(1) to the image gray processing after the car plate correction, comprehensive utilization OTSU automatic threshold, segmented areas growth and threshold value progressively improve three kinds of methods and extract the character prospects;

(2) utilize the priori of connected domain and character to generate the excellent character repertoire that cuts;

(3) the excellent wide height of all characters that cuts character repertoire is averaging, generates the plate template of standard;

(4) comprehensive utilization large-spacing and Template Information carry out cutting apart of character;

(5) utilize gradient and all segmentation results are estimated, choose gradient and maximum cutting apart as the optimum segmentation result.

Compared with the conventional method, Character segmentation algorithm disclosed in this invention, can process the situations such as uneven illumination is even, the inter-class variance distribution is the multimodal characteristic, the front background form and aspect mixing of character well.

Description of drawings

Fig. 1 is that the region growing Seed Points is selected process flow diagram;

Fig. 2 is segmented areas growth process flow diagram;

Fig. 3 is general flow chart.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, rather than whole embodiment.Based on the embodiment in the present invention, those of ordinary skills, not making under the creative work prerequisite the every other embodiment that obtains, belong to the scope of protection of the invention.

Embodiment is:

(1), to the car plate gray processing behind level correction and vertical school, utilize the OTSU automatic threshold algorithm to extract prospect;

(2) utilize the priori (the ratio of width to height etc.) of connected domain and character to generate the excellent character repertoire that cuts;

(3) if the excellent character number that cuts character repertoire, less than 2, utilizes segmented areas growth extraction prospect, the excellent character repertoire that cuts is supplemented;

(4) if the excellent character number that cuts character repertoire between 2 to 5, utilizes progressively raising method of threshold value to extract prospect, the excellent character that cuts is supplemented;

(5) the excellent wide height of character that cuts character repertoire is averaged, generates standard character, then, according to the prioris such as specification of car plate, generate the car plate standard form;

(6) utilize standard form information, find the position of intercharacter large-spacing, utilize large-spacing to cut apart to the left and right respectively, proofread and correct with the excellent character that cuts character repertoire simultaneously;

(7) if do not find large-spacing, the excellent character opsition dependent that cuts in character repertoire is sorted, choose the excellent of centre position and cut character and suppose that successively this character is the 1st to the 7th character of car plate, cut apart by the template that generates;

(8) to all segmentation results by gradient and estimating, choose gradient and maximum cutting apart as final segmentation result.

Claims

1. algorithm of the character cutting based on connected domain and template, it is characterized in that: at first, to the car plate gray processing after level correction and vertical correction, utilize the OTSU automatic threshold algorithm to extract the characters on license plate prospect, utilize the priori of connected domain and character to extract character, join excellent cutting in character repertoire, if the excellent number of characters that cuts is less than 2, utilize the segmented areas growth algorithm to extract the character prospect, the character that newly is partitioned into is joined excellent cutting in character repertoire, if the excellent number of characters that cuts is between 2 to 5, adopt progressively raising method of threshold value to extract the character prospect, the character that will newly be partitioned into equally joins excellent cutting in character repertoire, then all excellent wide height of character that cuts are averaging, use the wide height of this wide height as standard character, generate the car plate standard form, utilize the position of standard form information searching to the intercharacter large-spacing, utilize large-spacing to carry out to the left and right respectively Character segmentation, if do not find large-spacing, the excellent character opsition dependent that cuts in character repertoire is sorted, choosing the excellent of centre position cuts character and supposes that successively this character is the 1st to the 7th character of car plate, cut apart by the template that generates, finally to all segmentation results by gradient and estimating, choose gradient and maximum cutting apart as the optimum segmentation result.

2. a kind of algorithm of character cutting based on connected domain and template according to claim 1, it is characterized in that: comprehensive utilization OTSU threshold value, segmented areas growth and threshold value progressively improve three kinds of methods and extract the character prospect, and utilize the priori of connected domain and character to generate the excellent character repertoire that cuts.

3. a kind of algorithm of character cutting based on connected domain and template according to claim 1, it is characterized in that: the excellent wide height of all characters that cuts in character repertoire is averaging, the plate template of generation standard, and comprehensive utilization large-spacing and Template Information carry out cutting apart of character, utilize gradient and all segmentation results are estimated, choosing gradient and maximum cutting apart as the optimum segmentation result.