JPS62103782A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve a recognizing ratio to a similar stroke character such as HIRAGANA (Japanese syllabary) character by using the loop detecting result and determining the final recognizing result out of plural candidate category when the recognizing object character has many similar characters like the HIRAGANA character. CONSTITUTION:When a loop detecting part 17 receives the control signal from a recognizing processing part 15, the part reads the data of segments S1-S8 to constitute the stroke from a stroke buffer 14. The loop detecting part 17 examines the crossing relation between segments away over the constant value or above and decides the presence and the absence of the loop. In such a case, the loop detecting part 17 considers the imperfection of the loop due to the habit, etc., of the writer, for example, at the terminal point P9 of the stroke, a virtual segment S9 of the same direction and the same length as the final segment S8 is assumed and with the virtual segment S9 as the final segment, the presence and the absence of the loop are decided. The recognizing processing part 15 determines the final recognizing result based upon the category pair of the obtained candidate category and the deciding result of the presence and the absence of the loop.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to character recognition devices, particularly those using tablet human power devices.

[Technical background of the invention and its problems] Conventionally, there is an online character recognition device that uses a tablet input device. This character recognition device is configured to convert characters written on a tablet with a writing pen into a coordinate data string, and to recognize a pattern made of this coordinate data string based on a pre-stored standard pattern.

By the way, if the target characters include hiragana characters, etc., there will be many similar characters, making it difficult to determine the final recognition result in the recognition process. For this reason, there is a method for determining similar characters by detecting the presence or absence of loops in a pattern made up of strokes when writing a character.

By detecting this loop, it is possible to recognize mutually similar characters such as "ne" and "wa", "ro" and [ruj, for example.

Various methods have been developed for loop detection.

For example, as a first method, as shown in FIG. 4(a),
After sampling each coordinate of one stroke to obtain coordinate point series P1 to P15, all distances between coordinate points are determined for all sets of two coordinate points separated by a certain value or more. Coordinate point P3, which is the smallest distance among these distances. Distance li between P13
If d is smaller than a predetermined threshold, it is determined that there is a loop. However, in this first method, as shown in FIG.
), when the sampling of each coordinate point of one stroke is relatively coarse, two points P2 . The distance between P8 may be greater than a predetermined threshold. For this reason,
Even though there is a loop, it is determined that there is no loop.

In addition, as a second method, as shown in FIG. 5(a),
There is a method of determining angles e1 to e6 formed by adjacent segments in a pattern in which one stroke is composed of segment series 81 to S7. This second method is
This method determines that a loop exists if the absolute value e of the sum of the calculated angles is larger than a certain value (2π). However, in this second method, as shown in FIG. 5(b), the absolute 1111e becomes larger than a constant value (2π), and it may be determined that there is a loop even though there is no loop. be.

Roughly speaking, as a third method, as shown in FIGS. 6(a) to (C), there is a method of examining the intersecting relationship between both segments from the coordinates of their starting and ending points for all pairs of segments that are separated by a certain value or more. be. In this third method, segment pairs (Si, Sj
) is detected, it is determined that there is a loop. in this case,
As shown in FIG. 6(d), since there is no intersecting, T-shaped, or L-shaped connection relationship, it is determined that there is no loop.

However, when actually writing characters, a portion where a loop should originally be formed may separate due to the scribe's habits or the like, and it may be determined that there is no loop.

Therefore, even if the presence or absence of a loop is detected using the first to third loop detection methods, there is a problem in that the recognition accuracy of the character recognition device is not so high.

[Object of the Invention] An object of the present invention is to provide a character recognition device that uses a loop detection method and can improve the recognition rate for characters with many similar characters, such as hiragana characters.

[Summary of the Invention] The present invention provides a character recognition device equipped with a tablet input means, which includes a memory means storing a plurality of category pairs consisting of two similar characters, distinguished in advance by the presence or absence of loops, a loop detection means, and a recognition determination device. have the means. When the recognition result of the character recognition means consists of a plurality of candidate characters and includes a category pair stored in the memory means, the recognition determination means determines the likelihood of each candidate character corresponding to the detection result of the loop detection means and the category pair. Based on the recognition result, the final recognition result is determined from the recognition result.

With the character H1 device having such a configuration, it is possible to improve the recognition rate for similar characters that can be distinguished by the presence or absence of loops by using the loop detection results.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram showing the configuration of a character recognition device according to an embodiment of the present invention. In FIG. 1, a tablet 10
When a character is written with a writing pen, it outputs an electric signal according to the contact position coordinates of the writing pen. The tablet control unit 11 converts the electric @ sign output from the tablet 10 into a coordinate data string consisting of a digital signal. The coordinate buffer 7 memory 12 stores the coordinate data string output from the tablet control section 11.

The feature/stroke extraction unit 13 uses the coordinate buffer memory 1
A characteristic pattern and a stroke are extracted from the coordinate data string stored in 2. Stroke buff? Reference numeral 14 denotes a memory that stores stroke data extracted by the feature/stroke extraction unit 13. The recognition processing unit 15 performs a matching process between the feature pattern output from the feature/stroke extraction unit 13 and the Rin quasi pattern stored in the dictionary memory 16 in advance, and selects candidate characters (candidate categories) with high likelihood. Output sequentially. On the other hand, the loop detection unit 11 detects, from the stroke data stored in the stroke buffer 14,
This is a circuit that determines the presence or absence of a loop.

In addition, the recognition processing unit 15 stores a table (FIG. 2) consisting of a plurality of category pairs based on two similar characters that have been distinguished in advance based on the presence or absence of loops.

When a category pair with a high likelihood exists among the plurality of candidate categories, the recognition processing section 15 determines the final recognition result from among the plurality of candidate categories based on the loop detection result of the loop detection section 17. .

The operation of this embodiment in a character recognition device configured like a mouth will be explained. First, when an operator writes characters on the tablet 10 with a writing pen, an electric signal is output from the tablet 10 according to the contact position of the writing pen. The tablet control unit 11 converts the electric signal from the tablet 10 into a digital signal, and outputs position coordinate data (n-1 bits) corresponding to the contact position of the writing pen and a pen status flag (
Outputs coordinate data consisting of 1 bit). The pen status flag is a flag that indicates whether or not the writing pen has touched the tablet 10. Here, n is a value determined from the size of the writing surface of the tablet 10 and the spatial resolution, and is usually a value of about 10 to 15.

The coordinate buffer memory 12 sequentially stores coordinate data strings output from the tablet control unit 11. The special @/stroke extraction unit 13 sequentially reads coordinate data from the coordinate buffer memory 12 and extracts a coordinate data string in which the pen status flag continues to be on as a coordinate data string for one stroke. The feature/stroke extraction section 13 obtains a feature pattern (feature data series) based on the extracted coordinate data string and outputs it to the recognition processing section 15 . at the same time,
The feature/stroke extraction unit 13 divides the coordinate data string for one stroke into a plurality of segments, and outputs stroke data consisting of start and end point coordinate data of each segment to the stroke buffer 14.

The recognition processing section 15 performs recognition processing on the characteristic pattern output from the special@/stroke extraction section 13 based on the standard pattern stored in the dictionary memory 16. The recognition processing unit 15 outputs a plurality of candidate categories through recognition processing. Here, the recognition processing unit 15 stores a table as shown in FIG. 2, for example. This table is made up of a plurality of category pairs of two similar characters that are distinguished by the presence or absence of loops, such as the hiragana characters "ne" and "wa" and "ru" and "ro." When the category pair in the table is present among the plurality of candidate categories, the recognition processing section 15 outputs a control signal instructing the loop detection section 17 to perform a loop detection operation. Further, if the category pair in the table does not exist among the plurality of candidate categories, the recognition processing unit 15 outputs the candidate category with the maximum likelihood as the final recognition result.

Upon receiving the control signal from the recognition processing section 15, the loop detection section 17 reads data of segments 81 to S8 forming a stroke as shown in FIG. 3 from the stroke buffer 14. The loop detection unit 17 examines the intersecting relationship between segments separated by a certain value or more, and determines the presence or absence of a loop. In this case, the loop detection unit 17 takes into consideration the incompleteness of the loop due to the scribe's habits, etc., and detects the final segment S8 at the end point P9 of the stroke, for example, as shown in FIG.
Assuming a virtual segment S9 in the same direction and the same length,
The presence or absence of a loop is determined using this virtual segment S9 as the final segment. As a result, even if a character that originally requires a loop is written as a character with an incomplete loop due to the scribe's habits or the like, it can be detected that there is a loop.

In this case, even if virtual segment S9 is assumed for a character that originally does not have a loop, it is rare for a loop to be formed.

The loop detection unit 17 outputs the determination result of the presence or absence of a loop to the recognition processing unit 15. The recognition processing unit 15 determines the final recognition result based on the obtained category pair of candidate categories and the determination result of the presence or absence of a loop.

That is, for example, the requested category pair is "ro" and "ru".
Suppose that the likelihood ratio between the former and the latter is α.

When α≧1 (likelihood of “ro”≧likelihood of “ru”), if α is larger than a predetermined threshold α1 (α1>1), the recognition processing unit 15 rejects the loop if there is a loop. If there is no ``ro'', it outputs ``ro'' as the final recognition result. Also, α≦α
In the case of 1, the recognition processing unit 15 recognizes "ru" if there is a loop,
If there is no loop, "ro" is output as the final recognition result.

Also, when α〈1 (likelihood of “ro” (likelihood of “ru”)),
If α is smaller than a predetermined threshold α2 (α2<1), the recognition processing unit 15 outputs “ru” as the final recognition result if there is a loop, and rejects if there is no loop. Also,
In the case of α≧α2, the recognition processing unit 15 outputs “ru” if there is a loop, and outputs “ro” if there is no loop as the final recognition result.

That is, if the determination result of the presence or absence of a loop matches the one with higher likelihood of the category pair, the recognition processing unit 15 takes that candidate category as the recognition result. In addition, the recognition processing unit 15 rejects when the comparison results are contradictory, and outputs the candidate category corresponding to the determination result of the loop detection unit 17 as the recognition result when the likelihood date α is smaller than the threshold value.

In this way, if a pair of similar characters with different loops exist as candidate categories in the recognition process, an appropriate candidate category can be determined as the final recognition result based on the loop detection results. In this case, in the loop detection process, the presence or absence of a loop can be reliably detected by a method using the virtual segment S9 even if an incomplete loop is formed due to the scribe's habits or the like. For this reason,
When category pairs exist, the accuracy of recognition processing using loop detection results can be improved.

[Effects of the Invention] As detailed above, according to the present invention, when the recognition target character has many similar characters such as hiragana characters, the loop detection result is used to select the final recognition from among multiple candidate categories. The results can be determined with certainty. Therefore, the recognition rate for characters with similar strokes such as hiragana characters can be greatly improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a character recognition device according to an embodiment of the present invention, and FIG. 2 is a diagram showing an example of a table of category pairs stored in advance in the recognition processing section of the embodiment.
FIG. 3 is a diagram for explaining the operation of the loop detection section of the same embodiment, and FIGS. 4 to 6 are diagrams for explaining the conventional loop detection operation, respectively. 10...Tablet, 11...Tablet control unit,
13... Feature/stroke extraction section, 15... Recognition processing section, 17... Loop detection section. Applicant's agent Patent attorney Takeshi Suzue Gutai 1 ;;

Claims (1)

[Claims] A tablet input means for converting a coordinate position corresponding to a written character into a coordinate data string, and a character recognition process based on the coordinate data string output from the tablet input means to output a recognition result. character recognition means that stores a plurality of category pairs consisting of two similar characters that are distinguished in advance by the presence or absence of loops; and a memory means that detects the presence or absence of loops based on the coordinate data string output from the tablet input means. a loop detection means for detecting the detection result of the loop detection means and each candidate character corresponding to the category pair when the recognition result of the character recognition means is composed of a plurality of candidate characters and includes a category pair stored in the memory means; 1. A character recognition device comprising: recognition determining means for determining a final recognition result from the recognition results based on the likelihood of the recognition result.