JPH0335400A - Method for recognizing use code of number plate - Google Patents

Abstract

PURPOSE:To perform stable recognition free from an influence of the form of a character, halation of the sunlight, or the noise by making groups of the character in accordance with character part rates and binarizing picture multilevel data to satisfy rates of respective groups and recognizing the character. CONSTITUTION:Groups are made of a character 1 by character part rates, and the character is binarized to satisfy respective rates of groups, and matching between character parts of binarized data 2 to 4 and dictionary data 5 to 7 is performed to detect a binarization level stable against the variance of data. Since processing data 2 to 4 are binarized by rates of character parts adapted to dictionaries 5 to 7 in this manner, stable recognition free from an influence of the form of the character, halation of the sunlight, or the like is possible.

Description

[Detailed Description of the Invention] Industrial Application Field The present invention relates to a license plate usage code recognition method that images a traveling vehicle with a television camera and processes the obtained still image to recognize the vehicle's license plate. It is something.

BACKGROUND ART In image processing, binarizing multivalued data with gradations and compressing the data is an extremely commonly used technique, and various techniques exist.

There is a method of adding all the data in the processing area, finding the average value, and using this as the binarization level.

When high-speed processing is required, time may be spent on addition operations depending on the processing data, but this can be handled by thinning out the processing data to some extent.

Usually, there is a method of creating a histogram of the data in the processing area and dividing it into four parts, the text part and the background part. Figure 2 shows a method for creating a histogram for character data and determining the 2[level]. In the figure, a shows processed data, and b shows a histogram of the data. There are two peaks in the histogram, the black peak is the text area, and the white peak is the background area. Based on this characteristic, the Iff level is L/(-・
The character part and the background part can be appropriately separated at the lines (-, - lines).

In this way, even with the conventional method described above, image multilevel data can be properly 2
The binarization level to be converted into a value can be detected.

Problems that the invention aims to solve However, the conventional method described above has the following problems.

When detecting the binarization level using the average value, the average value is affected by the character ratio of each character.
There are problems with using it as a value level. If the character area ratio is high, the average value will be closer to the black level, and conversely, if the character area ratio is lower, the average value will be closer to the white level. When these are converted into 12-valued repeps, characters with a high character part conversion rate are converted into 2M fine characters, and characters with a low character part conversion rate are converted into 2M thick characters. It is not desirable that the shape of the character affects the level in this way. It is also easily affected by solar halation and noise.

On the other hand, when using a histogram, while there is no problem mentioned above, there is the problem of how wide the gradation width should be when creating the histogram. If the gradation width is not set appropriately,
Characteristics of the data cannot be clearly detected.

However, since the characteristics of an image vary greatly depending on the illuminance of the imaging state, even if a histogram is created with a certain fixed gradation width, the characteristics of the image may not be detected.

Further, depending on the image, it may not be possible to separate the text portion and the background portion, and in this case, it is not possible to derive an appropriate binarization level.

The present invention solves these conventional problems,
This provides an excellent method for detecting an appropriate binarization level.

Means for Solving the Problems In order to achieve the above object, the present invention divides characters into groups according to character portion ratios, and binarizes each group by matching the ratios.

Therefore, according to the present invention, by converting the ratio of the character parts into 2, it is possible to match the character parts between the binarized data and the dictionary data, and to make the character part consistent with respect to data fluctuations.
This has the effect that the value level can be detected.

Embodiment FIG. 1 shows the data structure of an embodiment of the present invention. In Figure 1, 1 is input multi-value data, 2-
4 is binary data that is binarized with each character part ratio, 6
7 are dictionary data of respective character portion ratios.

Next, the operation of the above embodiment will be explained. In the above example, when multi-value data is input, the ratio of the character part is 3.
Binarize to 1%, 40%, and 51% to create three binary data. These binary data have a ratio of 31 and a ratio of 40 among the dictionary data.
Pattern matching processing is performed on the data of % with the data of dictionary data group ■, and with the data of 61% with the data of group ■. Group ■ is composed of dictionary patterns with a 31% character portion for characters with few character portions (18 characters such as I, TS, etc.). Guμm pu■ is based on a dictionary pattern with a 40% character portion, and is composed of 48 intermediate characters.
m b■ is based on the dictionary pattern of character part 61 Hajime), character with many character parts (a).

It consists of 48 pieces (nu, ha, etc.). As a result of pattern matching, the currently processed data is binarized at an appropriate binarization level when matching with the dictionary group in which the character is placed, so that it can be correctly recognized.

In this way, according to the above embodiment, since the processed data is binarized with the ratio of the character parts matched to the dictionary, stable recognition is achieved without being affected by the shape of the characters or the effects of sunlight halation, etc. It has the effect of being able to do the following.

Effects of the Invention As is clear from the above embodiments, the present invention determines the binarization level during the binarization process of multivalued data by matching it with the character portion ratio of the dictionary. This has the effect of increasing the accuracy of pattern matching. Furthermore, since it does not depend on the level of multivalued data, it also has the effect of being resistant to the effects of changes in external illumination.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a code structure of a usage code recognition section according to an embodiment of the present invention, and FIG. 2 is a schematic explanatory diagram of binarized repeat detection using a conventional histogram.

Claims (1)

[Claims] A license plate usage code recognition method that divides the license plate usage code into groups according to the ratio of character parts, and binarizes image multivalued data to match the ratio of the groups.