JPS59186074A - Method for recognizing character with on-line system - Google Patents

Abstract

PURPOSE:To attain the recognition of an input character independingly of the order of a stroke by providing a character characteristic dictionary and a character characteristic collating section collating input character characteristic information and character characteristic dictionary information of the standard pattern. CONSTITUTION:A character coordinate data inputted from a tablet 1 is outputted to a rough classification discriminating section 3 and a stroke number collating section 5 via a stroke data extracting section 2. The discriminating section 3 decides a candidate character of the standard pattern to be collated and the character is outputted to a stroke number dictionary 4 and a character characteristic dictionary section 6. The dictionary 4 outputs the stroke number deciding information of the candidate character to the collating section 5. The collating section collates unknown input character stroke information and the stroke number deciding information so as to decide the stroke number of the input character and outputs it to a character characteristic collating section 7. The collating section 7 collates the candidate character characteristic with the unknown input character characteristic, the degree of similarity is calculated and the result is stored in a candidate character similarity register. Then, the character having the maximum similarity is taken as the result of recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a recognition method that can recognize written characters input in an online character recognition device used as an input device of an information processing device regardless of the stroke order.

(Background Art) Conventionally, in an online character recognition method, pattern matching is performed between an unknown input character and a standard pattern using information on the stroke order and number of strokes of a written input character.
Based on the result of the distance calculation, the one with the smallest distance can be used as the recognition result, or the feature indicating the stroke shape of the unknown input character (hereinafter referred to as stroke feature), the feature indicating the positional relationship between strokes (hereinafter referred to as the distance between strokes), etc. Recognition methods include extracting features (referred to as features), calculating similarities between these features and feature information of standard patterns stored in a character feature dictionary, and using the one with the maximum similarity as the recognition result. It has been taken.

For example, taking the number "4" as an example, first, the two-stroke standard pattern group in the character feature dictionary is determined as the standard pattern 11 to be matched with the input character using the stroke count information of the input character. Next, using the stroke order information of the input characters, the shape of the first human stroke 'L'' and the first stroke of the standard pattern is compared, and the shape of the second human stroke 'L'' is then compared.
The shape of 1゛ and the second stroke of the standard pattern is compared, and the positional relationship between the 11th and 2nd strokes is also compared, and the sum of these matching results is calculated as the similarity between the input character 11411 and the standard pattern. degree. Repeat the above operation for all standard patterns of 2 strokes9.
The one with the maximum similarity between zero response sentences is detected and used as a recognized character. Therefore, if input character II 4 +1
If it is written in the stroke order rL / 11, 11 L'', it will not be able to match it with the standard pattern correctly, resulting in erroneous recognition. As a pattern, the first stroke is /", the second stroke is 11 f, "4"
The degree of similarity is calculated for each and every combination of strokes, such as by using the stroke order information between the strokes of the standard pattern and the strokes of the input characters, and then applying the maximum degree of similarity to each stroke of the standard pattern. By matching the strokes of input characters, a recognition method that does not depend on stroke order is adopted.

However, with these methods, as the number of images increases, the number of standard patterns increases, and matching with standard patterns requires a large amount of time. It had some drawbacks.

(Problem of the Invention) In order to eliminate the above-mentioned drawbacks, the present invention provides a system that has a simple device configuration and can recognize input characters regardless of stroke order. First, after extracting the strokes of the written input characters, using character information unrelated to stroke order such as the number of strokes,
The recognized characters are roughly classified, and the input characters are further classified.
A stroke number dictionary that has stroke number determination information for each character whose stroke number can be determined by associating the stroke number with the stroke number of the standard pattern to be matched, and a stroke number dictionary that matches the stroke information of the input character with the stroke number dictionary information. Using the number matching section, determine the stroke number of the input character, and check the ! of that stroke. 1+i characteristic information and stroke 4-stroke interval characteristic information are extracted and used as character characteristic information of the unknown input character, and the character characteristic information of the standard pattern to be matched is stored! By having a character feature matching section that matches the input character feature information and the standard ◆ pattern characters/' Samurai feature information: i If information, it is possible to identify the same characters with different stroke orders Even if the information of multiple patterns is the same character, the character 7t1. Recognition processing can be performed without having it in the signature dictionary;
, the input characters can be recognized regardless of the order, and the device configuration is
'] is simplified and furthermore, the recognition processing time can be greatly shortened, and will be explained in detail below.

(Explanation of I'ttf configuration and operation/effect of 2n1 embodiment) FIG. 1 is a block diagram showing an embodiment of the present invention, in which 1 is a tablet, and 2 is input from coordinate data from the tablet 1. Stroke extraction of characters, and 0. Stroke data extraction unit for extracting features of stroke data, 6
4 is a large classification discriminator that uses character information such as the number of strokes that is highly dependent on the stroke order to roughly classify the recognized characters and determines the standard pattern character group to be matched. 5 is the stroke number dictionary in which stroke number determination information is stored for each character to assign a loaf number, and 5 is the stroke number dictionary information of the stroke number dictionary 4 and the stroke of the input character. Stroke number verification unit 6
7 is a character feature dictionary in which character feature information of each character is stored, and 7 is a character feature matching unit that compares and matches the character feature information of the character feature dictionary 6 with the feature information of the input character.

In the embodiment with the above configuration, the coordinate data of a written input character inputted from the tablet 1 is subjected to known preprocessing such as noise removal, smoothing, and normalization, and then sent to the stroke data extraction unit 2. sent,
Known stroke extraction is performed based on the preprocessed character information in the input order, and feature extraction of the strokes is performed, and these input character information are output to the major classification discriminator 3 and the stroke number collation unit 5. .

Based on the stroke data extracted by the stroke data extraction unit 2, the major classification discrimination unit 3 uses simple information such as the number of strokes that is not related to stroke order to select a group of characters (candidate character group) of a standard pattern to be matched. The recognition characters to be determined are classified into major categories, and these candidate character groups are sequentially stored in the stroke number dictionary 4.
, and the character feature dictionary section, and the stroke number dictionary section 4 outputs the stroke number determination information of the candidate character to the stroke number matching section 5 according to the input candidate character information. The stroke number matching unit 5 matches the stroke information of the unknown input character extracted by the stroke data extraction unit 2 with the stroke number determination information of the candidate character output from the stroke number dictionary 4, and calculates the stroke number of each input character. The stroke number of the input character is determined and output to the character feature matching unit 7, regardless of the stroke order of the input character, so that each stroke of the standard pattern of the candidate character is associated with the stroke number. The character feature matching unit 7 compares candidate character features consisting of stroke features and inter-stroke features of the standard pattern of candidate characters output from the character feature dictionary 6 according to the candidate characters output from the major classification discrimination unit 3. , the stroke characteristics of the input character and the inter-stroke characteristics are compared with the unknown input character characteristics using a method such as a known stroke analysis method based on the stroke number of the input character determined by the stroke number verification unit 5. , calculates the degree of similarity, and stores this calculation result in a candidate character similarity degree register (not shown).

In this way, all the candidate characters sequentially output by the major classification discriminator 3 are compared, and the character with the maximum similarity between the written unknown input character and the standard pattern of the candidate character group is determined as the recognition result. .

Below, regarding the online character recognition method according to the present invention, the three-stroke character "A" as shown in FIG.
711.1"-", will be described in more detail using an example of the stroke number dictionary shown in FIG. 3.

Input character coordinate data from the tablet 1 is sent to the stroke data extraction unit 2, where it is subjected to the above-mentioned known preprocessing, and then stroke extraction and stroke feature extraction are performed. The first manual stroke is "\", then the second manual stroke is "/", and the fifth manual stroke is "-1".

The major classification discrimination unit 3 uses information such as the number of strokes to determine candidate characters of the standard pattern to be matched. . . , etc. are sequentially output as candidate characters for comparison to the stroke number dictionary 4 and the character feature dictionary section 6.The stroke number dictionary 4 inputs input characters as shown in FIG. 2, an example of which is shown in FIG. The starting point coordinates of the n-th manual stroke (X
ns, Yns), end point coordinates (XnTYrrr)
Stroke number determination information that allows correspondence to be established by comparing the input stroke of an input character with the standard pattern stroke of the candidate character to be matched using a simple method such as determining the size of the coordinate value. is stored for each candidate character. Further, the stroke number of the standard pattern is determined so that the stroke number of the standard pattern of each candidate character can be determined uniformly, as shown in FIG. 6, regardless of the normal order of strokes. When "A" is input into the stroke number dictionary 4 as a candidate character to be matched, the first, second, and second characters of the input characters shown in FIG. 3rd manual stroke, ′\”, “/11. II-" is compared with the stroke number information of IIA" in the stroke number dictionary 4, an example of which is shown in FIG. 3, in the stroke number matching section 5.

As a result, the standard pattern for the candidate character "A" is the first stroke is "-〇, the second stroke is \゛, and the third stroke is o -
Since the stroke is "'/" and the first stroke information of this standard pattern is "YnT is the largest stroke among the three strokes", "-1" corresponds to the stroke of input character 11 A ii. The stroke number 1 is assigned to the stroke 111 (third manual stroke) of the input character "A". Similarly, the second stroke information of the standard pattern is the one with the larger Xn7 among the remaining strokes, so the input The stroke “\” of the letter “A” (
The stroke number 2 is given to the first manual stroke), and the input character “A” is added as the remaining stroke.
A stroke number 3 is assigned to the stroke "/" (second manual stroke).

The input character features constituted by the stroke feature information of the n-th stroke mill determined in this manner and the inter-stroke feature information are output to the character feature matching section 7. In addition, the character feature dictionary 6 stores candidate characters "
The character feature information of the standard pattern "A" is output to the character feature matching unit 7.The character feature matching unit 7 matches the input character feature information with the character feature information of the candidate character It A I+. , the similarity is calculated.The resulting similarity is very low.For the next candidate character ``E'', the candidate character ``E'' in the stroke number dictionary of 4, an example of which is shown in FIG. Using the stroke number determination information of '・, the first stroke "-" of the standard pattern °゛E''' is the largest stroke of Yn7 among the three strokes, and the second stroke "-" is the largest stroke of YnT among the remaining 92 strokes. Since the larger one and the third stroke "L" are the remaining strokes, each stroke of the input character "A'" is 1111
(Third manual stroke), "/" (Second manual stroke), '\" (First manual stroke) 1.2.
A stroke number of 5 is assigned, and the input character “A″′
Each stroke of ``-'' K /II, ``\°'' is a standard turn ``E'' of ``-'', ``l 11. A stroke number (nth stroke) is determined and compared to correspond to "L"', and the degree of similarity between the input character and the candidate character "E" is calculated.

As a result, there is a clear difference between the second and third strokes, so the degree of similarity is small. The above operation is repeated for the remaining candidate characters after "F" to calculate the degree of similarity of each candidate character, and as a result, "A" is determined as the character with the highest degree of similarity.

As explained in detail above, since the stroke of the unknown input character and the stroke of the standard pattern of the candidate character can be easily matched, recognition processing can be performed regardless of the stroke order of the input character.
Even if the same character is written in different stroke orders, there is no need to have multiple standard patterns of the same character in the surname feature dictionary, and it has the advantage of being able to recognize input characters no matter what stroke order they are written in. . In addition, according to this method, by having the stroke number dictionary 4 and the stroke number matching section 5, it is possible to compare the correspondence between input strokes and strokes of the standard pattern, as has been conventionally done. It is very simple because it eliminates the need to perform stroke matching by considering as many combinations as possible permutations of all strokes, or by changing the combinations by adding conditions to some strokes. Furthermore, since the strokes of the input characters can correspond to the strokes of the standard pattern of the candidate characters, there are advantages that the recognition processing time is extremely short and the device configuration can be extremely simplified.

In addition, in the above embodiment, when ranking strokes, the start point coordinates and end point coordinates are taken into consideration, but the center point coordinates, the rightmost (leftmost) coordinates, the uppermost (downward) coordinates, etc. are also included. can be set.

In carrying out the present invention, it is important that the input characters and candidate characters are ranked according to the same calculation rule, which uniquely identifies the corresponding strokes of the input character and candidate character. It is guaranteed that the comparisons can be made with high similarity and with a high degree of similarity. This is also important when comparing and matching the positional relationship between strokes, and the ranking determines which two strokes of the candidate character the two strokes of the target input character correspond to. can.

(Effects of the Invention) As is clear from the above description, the present invention can recognize input characters without being constrained by the stroke order of input characters in online character recognition where stroke order information of input characters is used as important information. This provides an effective means for shortening the recognition processing time, reducing the size of character feature dictionary information, and simplifying the device configuration.

[Brief Description of the Drawings] Figure 1 is a block diagram of an embodiment of the present invention, Figure 2 is a diagram showing an example of the input stroke order of a 3-stroke input character, and Figure 3 is an example of the contents of a 4-stroke number dictionary. This shows that. DESCRIPTION OF SYMBOLS 1... Tablet, 2... Stroke data extraction part, 3... Main classification discrimination part, 4... A-ri number dictionary, 5... Stroke number collation part, 6... Character Feature dictionary, 7...Character feature matching unit, Patent applicant Oki Electric Industry Co., Ltd. Patent application agent Megumi Yamamoto -

Claims (1)

[Claims] In an online character recognition method that performs character recognition by comparing and matching the character pattern of input characters written on a tablet with the character pattern of standard characters stored in advance in the dictionary section based on the stroke shape and the position between strokes, the character is recognized. A ranking is assigned to the constituent characters, and this ranking and an operation rule for deriving the ranking are stored in the dictionary section in association with each character, and when the character is written on the tablet, Discrimination information on the number of strokes that does not depend on the stroke order is extracted from the input character, and based on this discrimination information, a group of characters close to the input character are extracted from the dictionary section to form a candidate character group, and the calculation rules for each candidate character are used to calculate the number of strokes. An online character recognition method characterized by assigning a rank to each stroke of an input character, and setting strokes of the same rank between the input character and a candidate character as strokes to be compared.