JPH0281282A - Character recognition method - Google Patents

Abstract

PURPOSE:To limit candidate characters to stably recognize a character at a high speed by collating a density vector of the character expressing the pattern of the character with contents of a dictionary, where characters are expressed with the same form as this density vector, in a vector space. CONSTITUTION:When one character is divided to mXn-number of meshes and the density of each mesh is sampled, this character is expressed with an mXn- dimensional vector. This mXn-dimensional character vector is collated with dictionary vectors mXn-dimensionally expressed in the same manner to discriminate the character. When both mXn-dimensional vectors are different in one element (the density of one of mXn-number of areas) or several elements (densities of several areas) but are equal in the other elements at the time of collation between mXn-dimensional vectors, it is judged that different elements are caused by break or blur of the character, and the pertinent character can be taken as a candidate. Since the character has the same inclination of the variable density as a dictionary character though having a low contrast when these characters coincide with each other, correct candidate characters are limited. Thus, the character is stably recognized at a high seed.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a method for recognizing printed characters using image processing.
In particular, the characters la that can be recognized even when printed characters are cut or faded! Regarding the organ method.

[Conventional technology]

Conventional character recognition methods, such as Japanese Patent Application Laid-Open No. 61-255488
The letter i written in the issue bulletin! The identification method divides a character pattern into a plurality of regions, and compares and verifies the degree of similarity for each division unit with a dictionary pattern to make a determination.

[Problem to be solved by the invention]

The conventional techniques mainly apply pattern matching methods, and in any case, most of them are recognition algorithms using binary images. Therefore, when applied to the recognition of characters printed on high-quality paper, etc., good matching results can be obtained, but if the printed characters are cut, scratched, or have poor contrast, a good binary image cannot be obtained. In order to determine this, the recognition algorithm is complicated), and there is a problem that the recognition time increases, the false recognition rate increases, and the number of rejects increases.

An object of the present invention is to provide a character recognition method capable of recognizing printed characters with poor contrast due to cuts or scratches in a good manner and at high speed.

[Means to solve the problem]

The above task involves cutting out one character from an input image, dividing the character into vertical and horizontal meshes, sequentially sampling the representative densities within each mesh, and expressing the character pattern as a density vector. This is achieved by comparing this density vector with a dictionary that represents characters in the same format on the Ventre space, and limiting candidate characters.

[Effect]

When one character is divided into mXn meshes and the density within each mesh is sampled, this one character is expressed as an mXn-dimensional vector. Characters are determined by comparing this mxn-dimensional character vector with the same dictionary vector expressed in mXn dimensions. ) or the number element (density of the number area) is different, but if the other elements are the same, it can be determined that the different element is caused by cut or scratched characters, and legitimate characters can be used as candidates. Characters with poor contrast have a higher density than characters with good contrast, which is 9mX.
The values of the n elements are different. However, the trends of the values of mXn elements and the trends of shading show similar trends if they are the same character. Therefore, in this case as well, valid candidate characters can be limited.

〔Example〕

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

FIG. 2 is a block diagram of a character recognition system to which a character recognition method according to an embodiment of the present invention is applied.

The synutem is for reading printed characters on the surface of a workpiece 21, which is a product on which characters are printed and a product on which the product is mounted. Based on a command from the image processing device 20 via the mechanical control interface 22, the chiffle controller 25 controls the X table 24. The X table 25 is controlled so that the work 21 comes within the field of view of the TV left camera 6. The work 21 that has entered the field of view is uniformly illuminated by the illumination light source 27, and an image of the surface is captured by the TV left camera 6 into the image processing device 20.

The image processing device 20 has a hardware configuration shown in FIG. The image processing device 20 includes an image input/output circuit 51 that inputs signals from the TV left camera 6 and outputs them to a monitor TV 28 (FIG. 2), an image memory 52 for storing input image data, and mechanical control. A general-purpose input/output circuit 63 that inputs and outputs external signals such as the interface 22, an arithmetic circuit 66 that performs operations such as inter-image operations and evaluation programs, and a program memory 65 that stores programs and data.
CPU 54, which controls the entire operation mentioned above. It also includes a multipath 67 for signal communication between these circuits.

Reading of printed characters is performed by performing processing according to the flow shown in FIG. 1 inside the image processing device 2Q. This processing program is stored in the program memory 35. First, using the image input/output circuit 51, an image of a character string drawn on the surface of the workpiece 21, which is already within the field of view of the TV left camera 6, is taken into the image memory 32 (step 11
). Then, the following processing is performed by the arithmetic circuit 3. First, cut out a rectangular area in which each character exists from the image (step 12).The cutting method at this time is as follows.
Markers for cutting out are printed at the beginning and end of the character string in advance, and the position of the characters is detected using the markers. Furthermore, the extraction of each character is performed by utilizing the fact that the pitch of the characters is constant. Then, the cut out character area is divided into areas of length and width of mXn, and the character pattern is evaluated by calculating the similarity (step 13). Character candidates are restored by extracting edges for each mXn region, and as a result, if there are multiple candidate characters, the recognition result is output (step 16), and if there is only one candidate character, the recognition result is output (step 16). 14) Do. (cut) The above processing (steps 13 to 16) is performed on all the characters that appear, and the reading of the characters in the input image is completed.

Character evaluation in step 13 is performed according to the flow shown in Fig.
Divide into Xn areas (step 41). Thereafter, the density representing each area is sampled within the area, and the input cover pattern is expressed by a density vector having this density as a component (step 42).

For example, m=5. If n=3, it becomes a 15-dimensional vector. Next, the degree of similarity for identification is calculated between the density vector of the input pattern and the pattern vector in the dictionary (step 45). At this time, if each component of the density vector all matches each component of the pattern vector in the dictionary, the input character can be determined reliably, but for example, if the character line is broken or blurred, the components of the density vector If there are some areas that do not match, it cannot be determined based on the similarity alone. Therefore, a threshold value for the degree of similarity is set, and using this predetermined threshold value, all characters with a high degree of similarity are extracted, and these are all used as candidate characters (step 44).

If a plurality of candidate characters are selected as a result of the similarity calculation in step 15, character evaluation is performed in accordance with step 70- in FIG. 5 to narrow down the candidates to one for the next sticker 5. If the candidates cannot be narrowed down to one, extract the edges of the character lines for each of the mXn divided regions (step 51), and extract the character line types in each of the mXn regions obtained as a result (see Figure 6). (only six types are exemplified) are compared with the dictionary i (step 52), and the matching candidate character is determined as the target character.

As a result of the above, it becomes possible to recognize characters that were previously impossible to recognize due to poor contrast, broken or blurred character lines, etc. In addition, an easy-to-recognize pattern is used in advance in the font design of the printed characters to be recognized. By doing so, reliability can be improved.

〔Effect of the invention〕

According to the present invention, when characters are printed in poor condition, for example,
High-speed and stable recognition can be achieved without being affected by images with poor contrast, broken character lines, or blurred images.

[Brief explanation of the drawing]

FIG. 1 is a processing flowchart of a character recognition method according to an embodiment of the present invention, FIG. 2 is a system configuration diagram of a character recognition device according to an embodiment of the present invention, and FIG. 3 is an image of the image shown in FIG. 4 is a detailed block diagram of the processing device; FIG. 4 is a 70-chart of recognition processing using similarity; FIG. FIG. 6 is a diagram showing the types of character lines in each region obtained by edge extraction. 13... Similarity evaluation step, 15... Edge extraction evaluation step, 42... Density value sampling step, 20... Image processing device, 21... Recognition target, 2
2...Interface, 23...Controller, 2
4...X table, 25...y table, 26...
- Camera, 27... Magnifying lens, 28... Monitor. Figure 1

Claims (1)

[Claims] 1. In a character recognition method that recognizes printed characters such as alphanumeric characters and symbols by image processing, one character is extracted from an input image, and then the character is divided into vertical and horizontal meshes, and each character is divided into vertical and horizontal meshes. Then, the representative densities within the mesh are sequentially sampled, the character pattern is represented by a density vector, and this density vector is compared with a dictionary that represents characters in the same format on the vector space. A character recognition method characterized by limiting characters. 2. In claim 1, when there are a plurality of candidate characters in the comparison with the dictionary pattern, an edge in each mesh is detected, and thereby it is determined what the input character is from among the candidates. A character recognition method featuring: