JPH0436431B2 - - Google Patents

Description

[Detailed Description of the Invention] (a) Field of Industrial Application The present invention relates to a method for recognizing handwritten or printed characters, and particularly to a method for recognizing uppercase and lowercase letters.

(b) Prior Art In recent years, with advances in signal processing technology and electronics, research on character recognition technology has progressed rapidly, and various OCRs have been developed. However, many of these types of devices have limitations on the characters that can be recognized, and there are almost no devices that can accurately recognize ordinary sentences. In particular, "A", "A", which have the same character shape but only the size of the letters,
There is nothing that can recognize uppercase and lowercase letters in katakana or hiragana, such as ``tsu'' and ``tsu''. As shown in the figure above, even if you input the lowercase letter ``yo'', the recognition result often resulted in the uppercase letter ``yo'' being output.

In addition, in order to improve the recognition rate, there is a method that normalizes the size at the preprocessing stage, but in this case, the character pattern after normalization is almost the same for uppercase and lowercase letters. As disclosed in Electronic Technology Research Institute Research Report No. 834, ``Research on Handwritten Kanji Recognition Using Cell Features,'' lowercase letters are currently not recognized.

(c) Purpose of the Invention The purpose of the present invention is to make it possible to recognize lowercase letters in Katakana or Hiragana, which differ only in letter size and have the same shape. This is a standard feature pattern that does not need to be specially prepared.

(d) Structure of the Invention The present invention has a first area that stores a unique standard feature pattern of characters that can be either uppercase or lowercase, and a second area that stores a standard feature pattern of characters that can only be uppercase. a dictionary section having a region, the input character pattern is normalized in terms of size, a feature pattern is extracted from the normalized character pattern, and the feature pattern is converted into a standard feature registered in the dictionary section. In a character recognition method that recognizes a character by matching it with a pattern, a detection means for detecting the size of the character of the input character pattern is provided, and when the input character is determined to be an uppercase character by the means, the matching range is determined. is defined as both the first and second areas, and when it is determined that it is a lowercase letter, the matching range is limited to the first area, and based on the matching result and the detection result of the detection means, the matching range is determined to be output. It is configured to determine the characters.

(E) Embodiment FIG. 1 is a block diagram showing the overall structure of a character recognition device for realizing an embodiment of the present invention, in which 1 is a character observation section, 2 is a preprocessing section, and 3 is a feature extraction section. section, 4 is a pattern matching section, 5 is a dictionary section,
6 is an identification section, and 7 is a character code memory.

As shown in FIG. 2, the characters to be recognized are handwritten on a manuscript paper 8 on which only the text entry frame is printed, and the handwritten characters are input into the character observation section 1. The character observation unit 1 scans and photoelectrically converts all input handwritten characters using a CCD image sensor or the like, converts each character into an observed character pattern according to the shade, and sends the pattern to the preprocessing unit 2. In the preprocessing unit 2, first, the binarization unit 21 converts the observed character pattern into an input character pattern in which each grid point of an M×N mesh takes “0” or “1”, and converts the character pattern to be recognized into 1. Cutting out only the characters, using the character cutting section 22,
Cuts out a character pattern surrounded by a circumscribed rectangular frame. That is, as shown in Fig. 3, in an input character pattern of M×N mesh, tentacles are extended from each side toward the center, and when a character part is first encountered, the circumscribing frame of that side is detected. The information of the portion surrounded by each circumscribed frame detected by the method is extracted as a character pattern. At this time, the coordinates x ₁ , x ₂ and y ₁ , y ₂ of each circumscribed frame are detected. The extracted character pattern is subjected to expansion/contraction processing in the normalization unit 24, and the size is first normalized, and then the position is also normalized. Therefore, for example, "a"
When uppercase and lowercase character patterns with the same character shapes as ``A'' and ``A'' are input, these character patterns after normalization will be the same. Then, a feature pattern is extracted from the normalized character pattern by a feature extraction unit 3.

By the way, the character size detection unit 23 calculates the horizontal length _a and the vertical length b of the characters of the input character pattern, respectively, based on the coordinate information detected at the time of character cutting.
x ₁ +1) and b=(y ₂ -y ₁ +1), and these values a and b are compared with predetermined thresholds W and H, respectively. When a≧W and b≧H, that is, when it is an uppercase letter, “0” is added as the dimension flag Fd, and when a<W and b<H, that is, when it is a lowercase letter, “1” is added as the dimension flag Fd. ” is added. Demension flag obtained in this way
Fd and the extracted feature pattern are input to the pattern matching section 4 and stored in buffers 41 and 42, respectively.

Here, to explain the dictionary section 5, uppercase and lowercase letters of katakana and hiragana, which differ only in size and have the same shape, such as "A" and "A",
Regarding the characters "tsu" and "tsu", "yo" and "yo", and "ya" and "ya", the same standard feature pattern is stored in the dictionary section 5, and specifically, As shown in FIG. 1, standard feature patterns of these characters that can be uppercase or lowercase are stored in the area 5a of addresses 0 to 8 of the dictionary section 5, and
~99 area 5b stores standard characteristic patterns of characters that can only be uppercase letters.

Next, the processing of the pattern matching section 4 will be explained.

First, the control section 43 sets the start address "0" of the dictionary section 5 in the address counter 44, and at the same time sets a different initial value in the character number pointer 45 depending on the contents of the dimension flag Fd. . That is, when the dimension flag Fd is "0", "0" is set, and when it is "1", "100" is set. Then, the standard feature pattern is read from the address 0 of the dictionary section 5 indicated by the address counter 44, and the distance calculation section 46
Pattern matching is performed with the characteristic pattern of the input character pattern stored in step 2, and the distance between the two patterns is calculated. The distance as a calculation result is sent to the identification unit 6 in pairs with the character number that is the content of the character number pointer 45. When matching with one standard feature pattern is completed, the address counter 44 and character number pointer 45 are sequentially incremented, and the standard characteristics stored in addresses "1", "2", "3", etc. following the first address "0" are Perform pattern matching with characteristic patterns sequentially,
The distance is sequentially output along with the character number. When the dimension flag Fd is "1", the control section 43 limits the range to be matched to the area 5a of the dictionary section 5, so that the contents of the address counter 44 reach the final address 8 of the area 5a. After that, the pattern matching process is finished, and when the dimension flag Fd is "0", the final address 99 of the dictionary section 5 is reached in order to perform matching with the standard feature pattern of the entire area of the dictionary section 5. The pattern matching process continues until Therefore, the buffer 61 of the identification unit 6 stores data consisting of 100 character numbers and distances when the dimension flag Fd is "0", and stores 9 data when it is "1". It will be done.

Here, the input character is a capital letter "a" with the same glyph shape.
In the case of the lowercase letter "a", these feature patterns are pattern matched with the same standard feature pattern stored in the area 5a of the dictionary section 5, and the calculated distances are approximately the same values α, β, γ...
...becomes. However, character numbers start from "0" for uppercase letters and "100" for lowercase letters, so the data (character number, distance) output to the identification unit 6 is α) (1, β) (2, γ)... For lower case letters (100, α) (101, β) (102, γ)...
becomes. The identification unit 6 compares the distances of all these data, identifies the one with the smallest distance, and searches the character code memory 7 based on the character number. The character code memory 7 stores character codes corresponding to character numbers, and stores character codes from 0 to 99.
For character numbers 100 to 108, uppercase character codes corresponding to the standard feature patterns stored at addresses 0 to 99 of the dictionary section 5 are stored in order. The lowercase character codes corresponding to the standard feature patterns stored in 8 to 8 are sequentially stored. Therefore, as a result of the identification by the identification unit 6, an uppercase character code or a lowercase character code with the minimum distance is output, depending on the character size of the input character pattern.

In this way, the character code output from the identification unit 6 is determined based not only on the distance as a calculation result of the distance calculation unit 46 but also on the dimension flag Fd as a detection result of the character size detection unit 23. Ru.

By the way, to give an example of the threshold values W and H, the manuscript entry frame is 13 mm square, and the character observation section 1 is
When the resolution of the CCD image sensor is 8 lines/mm, the M value and N value of the M×N mesh are "96", but as the threshold values W and H, for example, "32" may be used.

(F) Effects of the Invention The character recognition method of the present invention determines the character to be output based not only on the distance calculation result in pattern matching but also on the detection result of the character size of the input character pattern. It also becomes possible to recognize lowercase letters in katakana and hiragana, which differ only in letter size but have the same shape, making it possible to recognize general handwritten text. In addition, the standard feature pattern for lowercase letters can also be used as the standard feature pattern for uppercase letters, so the capacity of the dictionary section can be reduced.
Furthermore, since recognition of lowercase letters is achieved by a very simple process of detecting the size of the characters in the input character pattern, it is possible to downsize the system. Moreover, when the input characters are lowercase letters, the matching range is limited to the first area of the dictionary section, so that the processing speed is increased and the recognition rate is improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the overall structure of a character recognition device for realizing an embodiment of the present invention, FIG. 2 is a diagram showing the format of an input document, and FIG. 3 is an explanatory diagram showing an input character pattern. be. Explanation of main figure numbers, 1...Character observation section, 2...Preprocessing section, 3...Feature extraction section, 4...Pattern matching section, 5...Dictionary section, 6...Identification section, 7... …
Character code memory.

Claims (1)

[Claims] 1. A dictionary section having a first area storing a unique standard characteristic pattern of characters that can be both uppercase and lowercase letters, and a second area storing a standard characteristic pattern of characters that can only be uppercase letters, The input character pattern is normalized in terms of size, a feature pattern is extracted from the normalized character pattern, and the feature pattern is matched with a standard feature pattern registered in the dictionary section to perform character recognition. In the character recognition method performed, a detection means for detecting the character size of the input character pattern is provided, and when the input character is determined to be an uppercase character by the means, the matching range is changed to the first and second character sizes.
and when it is determined to be a lowercase letter, the matching range is limited to the first area, and the character to be output is determined based on the matching result and the detection result of the detection means. A character recognition method featuring