JPH0421080A - Character recognition device - Google Patents

Abstract

PURPOSE:To reduce erroneous recognition by deciding one character from analogous characters stored in a second storage means as a recognition result corresponding to position information stored in a first storage means. CONSTITUTION:An image is inputted from an image scanner 108, etc., and a character is segmented one by one from an input image by a CPU 100, and the circumscribing rectangle of a segmented character is found, and it is normalized and is expanded/reduced to fixed size. The feature of a character image normalized by the CPU 100 is extracted, and similarity is calculated by using a feature extracted by an identification calculation part 105 and an identification dictionary stored in a ROM 103, and it is decided whether or not the analogous character exists by using an analogous character table stored in the ROM 103. At this time, since the position information of the character when the segmentation of the character is performed is stored in memory 104, either the analogous characters is selected by the CPU 100 based on the position information of the character, and a recognition result is outputted to a CRT 106. Thereby, the erroneous recognition can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character recognition device that is capable of identifying similar characters and is included in characters to be recognized.

[Prior Art] Conventionally, character recognition devices have at least an identification dictionary of the number of characters to be recognized as an average feature of training data, and an input image is cut out for each character and , extract features, calculate similarity using an identification dictionary, and output the highest similarity as the recognition result, or output several recognition candidates in order of similarity to a display device or storage device. There was something to do.

[Problem to be solved by the invention] However, in the above conventional example, it is difficult to normalize the size of the cut out characters, but to normalize the size of the cut out characters, it is difficult to normalize the size of the cut out characters. It becomes difficult to distinguish between different characters, and since the character recognition dictionary treats each character as a different category, it may be difficult to distinguish between characters that are different from each other. There was a drawback that the result was a recognition result.

[Means for Solving the Problems] According to the present invention, in order to solve the above-mentioned drawbacks, there is provided a means for inputting image information, a cutting means for cutting out character information from the input image information, and a candidate for the extracted character information. an identification means for identifying characters; a first storage means for storing positional information of the extracted character information; a second storage means for storing a plurality of similar characters as one category; and the candidate character. determining means for determining whether or not a similar character exists in , and determining one character as a recognition result from the similar characters stored in the second storage means according to the position information stored in the first storage means. have the means.

[Embodiment 1] FIG. 1 is a diagram showing the basic configuration in an embodiment of the present invention, where 100 is a central processing unit (CPU) that performs calculations such as the flowcharts in FIGS. 8 and 9, and 101 is a character/symbol. Keyboards (K, B) are used to input text and give instructions for correcting erroneous recognition, 102 is a Voint Ink device PD, and 103 stores a dictionary used to recognize characters, etc. REIT ONLY MEMORY (ROM),
IO2 is a memory that stores data read by the scanner 108; 105 is an identification calculation unit that finds candidate words from the data read by the scanner 108 and calculates the degree of dissimilarity of each; 106 is a CRT; 107 is the scanner 108 interface (SCAN I/
F) 108 is a scanner that reads image information.

FIG. 2 is a block diagram of an embodiment that best represents the features of the present invention. In step 1, an image is input from an image scanner 108, etc., and in step 2, the CPU 100 processes an image from the input image.
Characters are cut out character by character, 3. CPU 100 calculates the circumscribed rectangle of the cut out characters, normalized and scaled to a certain size, 4. CPU 1[]0 normalizes the characteristics of the character image. 5 and the identification calculation unit 105
The similarity is calculated using the features extracted by 6 and the identification dictionary 9 stored in the ROM 103.
It is determined whether or not there are similar characters using a similar character table 11 stored in . The position information of the character when the character is extracted in step 2 is stored in the memory 104 in step 10, so based on this character position information, one of the similar characters is selected by the CPU 100 in step 8, and the CRT 105 is selected in step 7.
Output the recognition results to .

Here, according to the flowchart shown in FIG.
The flow of processing performed in step 100 will be explained in detail.

An image is input from the scanner 108 (Sl), a line is drawn from the input image information (S2), and the height h of the line as shown in FIG. 3 is determined (S3). Next, characters are cut out as shown in FIG. 3 (S4). Once the characters are cut out and the number of characters in that line is known, bits representing the lower character status are secured in the position information storage section in the memory 104 for the number of input characters (S5).

It is determined whether the character exists below the height of the top pixel hx- (S6), and if it exists below x%, the character is determined to be a lower character, and the position information is Turn on the bit of position information (S7), and turn off the bit of position information if it is not judged in S6 that it exists below (S7).
8). If there is the next character (S9), the process returns to S6, and if there is no next character (S9), the process moves to the next line. If there is a next line (SIO), return to S3 and repeat steps 83 to S10 until the last line. If there is no next line (SIO), normalize the circumscribed rectangle of the cut character image.
), character characteristics are depicted (512), and discrimination calculation is performed in the discrimination calculation unit 105 (313). The identification dictionary 9 used when performing the identification calculation here is a ROM as shown in FIG.
103, and two letters that are distinguished by size are assigned a category number (l to n) as an l category. Once the extracted characters have been identified and their similarity calculated (S13), it is determined whether the category with the highest degree of similarity exists in the similar characters Chiful (S14).

The similar character table 11 is stored in the ROM 103 as shown in FIG.
For each category number in which similar characters exist in the identification dictionary 9, the categories when the position information bit is on and when the position information bit is off are included.

To explain using the ts3 diagram as an example, let us assume that the result of the identification calculation section is fm and IJ in category 1 number. Further, the position information storage section is r off, on, and off J. Since the category number m is included in the similar character table 11, the similar character determination unit 6 determines that it contains similar characters, and the similar character selection unit 8
If the category number is m, the bit is turned off or recognized. The similar character determining unit 6 determines that there are no similar characters for the category number father, and the recognition result is ``g''. S14
If it is determined that the similar character does not exist in Chiful,
S] The category identified as the largest similar boat in step 3 is the recognition result. In addition, if it is determined in S14 that the similar character exists in the similar character table, then in S16 the memory 1 is further
It is determined whether the position information bit in 04 is on.

If it is determined in S16 that the bit is on, the lowercase letter in the category of the similar character table 11 is set as the recognition result, and if it is determined in S16 that the bit is not on,
The uppercase letters in the category of the similar character table 11 are taken as recognition results.

S15. 'S17. When the recognition result is displayed in S18, it is determined whether there is a next character or not (519), and if there is a next character, it returns to Sll and returns to the last character from Sll to S ],
9 is repeated, and if it is determined in S19 that there is no next character, the result is displayed on the CRT 106 (S20).

[Embodiment 2] FIG. 6 is a diagram illustrating another embodiment of the similar character selection method. In this example, when the characters to be recognized are alpha-heads and symbols, we will explain how to identify characters that have the same character characteristics but differ only in character position by taking the method of distinguishing between FJ and FyJ as an example. .

In this embodiment as well, the structure of character recognition as shown in FIGS. 1 and 2 is the same as in the first embodiment, so it will not be described here.

Here, according to the flowchart shown in FIG.
The flow of processing performed in step 100 will be explained in detail.

In the process in FIG. 9, S1-54S11-S1
3. S19. Since S20 is the same as in the first embodiment, it will not be described here.

Characters are cut out in the same manner as in Example 1 (S4), and the sixth
As shown in the figure, the length from the top of the clipping frame to the top of the character and the length from the bottom of the character to the bottom of the clipping frame are determined (S:11), and the position information is stored in the memory 104. 10 and compares the sizes of U and V (S32). If it is determined in S32 that u < v, in S5 the bit of the position information secured in S5 is turned on (533), and in S32 u>v If it is determined that the position information bit is turned off (S34). 535) to determine whether there is a character next, and if there is a character next,
Return to step 32 and repeat steps S32 to S35 until the last character. If S35 determines that there is no next character, 336
Proceed to and determine whether there is a next step. If there is a next row, the process returns to S3 and repeats S3 to 336 until the last row. If it is determined at 536 that there is no need to proceed to the next step, the process proceeds to Sll.

After performing the character identification calculation in S13, it is determined whether the category with the highest degree of similarity is category 2 (537)
, If it is determined that there are no 9 categories, the category with the highest degree of similarity is selected as the recognition result538),
If it is determined that it is in the category of 337, it is determined whether the position information bit in the memory 104 is on or not (539), and if it is determined that the bit is on, IrtJ is set as the recognition result. 540) If it is determined that the bit in S39 is not on, set J as the recognition result (S41).

In addition, by finding U and ■ in the same way, and identifying characters with the same characteristics but in different character positions based on the relationship between U and ■, a table like the one shown in Figure 7 can be set up. It goes without saying that other characters are also possible.

[Effects of the Invention] As explained above, according to the present invention, by selecting uppercase and lowercase letters after recognizing similar characters as one character, it becomes possible to reliably distinguish between similar characters. It seems to have the effect of reducing misrecognition.

As described above, according to the present invention, by combining similar character identification dictionaries into one, there is an effect of reducing dictionary capacity and an effect of increasing processing speed.

[Brief explanation of drawings]

Fig. 1 is a basic configuration diagram of an embodiment of the present invention, Fig. 2 is a block diagram of a character recognition device implementing the present invention, and Fig. 3 is an explanation of cutting out characters in an input image and retrieving position information. Figure, ts4 diagram is a diagram explaining the inside of the identification dictionary, Figure 5 is a diagram explaining the similar character table, and National Policy 6 diagram is an explanatory diagram of character extraction and position information extraction of the method for distinguishing between "Yu 1" and ", 1. Fig. 7 is a diagram showing the conditions for selecting similar characters, Fig. 8 is a flowchart showing the processing of the first embodiment, and Fig. 9 is a flowchart showing the processing of the second embodiment. The extraction unit 3, the normalization unit 4, the feature extraction unit 5, the identification calculation unit 6, the similar character determination unit 7, the recognition result output unit 8, the similar character selection unit 9, the identification dictionary 10, the position information storage unit 1, the similar character table 2 is the character cutting frame

Claims (1)

[Scope of Claims] A character recognition device comprising means for inputting image information, a cutting means for cutting out character information from the inputted image information, and an identification means for identifying candidate characters for the cut out character information. a first storage means for storing positional information of character information that has been selected; a second storage means for storing a plurality of similar characters as one category; a determination means for determining whether or not a similar character exists among the candidate characters; A character recognition device comprising a determining means for determining one character as a recognition result from among similar characters stored in the second storage means in accordance with position information stored in the first storage means.