JP3470541B2 - Online handwritten character recognition device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line handwritten character recognition device, and more particularly to a handwritten character recognition device which has a small dictionary size and can be recognized with high performance even for continuous input such as one-stroke writing. Regarding the device.

[0002]

2. Description of the Related Art Character recognition methods for an on-line handwritten character recognition device are classified from the input restriction method into (1) a method of fixing the number of strokes and the stroke order, (2) a method of freeing one, (3) There is a method to make free without either limitation. On the other hand, as a classification from the algorithm, (4)
There are a structure analysis method for analyzing the structure of strokes, and (5) a pattern matching method for pattern matching the stroke feature points with a dictionary. Each has its own characteristics, and it is decided in consideration of performance.

Generally, the method (1) or (2) is adopted from the viewpoint of the recognition time, and from the viewpoint of dictionary learning,
The method (5) is often used.

Conventionally, in order to deal with these continuous characters, a commonly used method is to register a standard pattern of continuous characters in addition to the standard pattern (dictionary) of regular writing.

As a method of not registering a continuous character standard pattern, as shown in JP-A-57-45679 and JP-A-8-249424, not only on-stroke but also off-stroke (stroke during pen-up) is performed. Virtually temporarily combine (make one stroke) and divide the stroke into equal parts, or
There is a method of extracting feature points at equal intervals and matching this with a dictionary.

On the other hand, the feature point correspondence method uses the dynamic programming (DP) method for feature points to match the input pattern and the dictionary pattern as described in the above-mentioned Japanese Patent Laid-Open No. 8-249424. , Method of calculating distance value, IEICE Technical Report, PRL 84-68
(1985.1.25), PP19-PP26, “Point extraction method for corresponding points between standard characters and consecutive characters in online character recognition”, focusing on off-stroke stability, the corresponding points with off-stroke There is a method of extracting and recognizing continuous characters.

In addition, the method of extracting the characteristic points of the stroke has hitherto been used in the IPSJ Journal Vol. 27, No. 5 (Showa 6).
May 1st), PP492-PP498, "Strokes, stroke order,
As described in "Online handwritten figure recognition method that does not depend on rotation and division", a method has been devised by using the area method that is not affected by camera shake, noise of the coordinate input device, and the like. As described in Japanese Patent Laid-Open No. 8-249424, there is an angle method in which a stroke is traced and a bending angle point of a predetermined angle or more is used as a feature point.

Further, in the matching process, generally, a detailed identification method for each category is used for a character group that cannot be recognized by any means. For example, IEICE Transactions, D-2, NO. 4, PP 592-599, "Design and application of handwriting input interface for multimedia terminals", stroke end direction ("re" and "wa"),
The recognition performance is improved by using 175 characteristic functions such as the stroke start point position (“” and “sword”) and the difference in the number of strokes (“” and “”). Furthermore, JP-A-55-
In Japanese Patent No. 33245, crossing of stroke pairs is detected,
It is described that they are distinguished (for example, "ni", "earth", and "top").

Recently, a personal digital assistant (PDA: Personal D
igtal Assisitant) has become popular, but generally, it requires a lot of recognition time when the stroke count and stroke order are free, and in order to make the recognition speed comfortable, the clock speed operates at the maximum speed of the system during character recognition. I am letting you.

[0010]

The above-described conventional method for recognizing continuous characters has the following problems.

First, when various continuous characters are registered in the standard pattern, the dictionary size becomes large, and the recognition speed and recognition rate decrease.

Secondly, in the method of generating approximate points including off-strokes, instead of not registering the continuous character in the dictionary, the off-stroke approximate points are also registered.
There is a problem that the dictionary size becomes large. Similarly,
The larger the dictionary size, the slower the recognition speed.

Third, the off-stroke corresponding point extraction method has a problem that recognition of continuous characters depends on off-stroke. Therefore, it is necessary to estimate the off-stroke of the input pattern from the off-stroke of the standard pattern, and there is a problem that the estimation error or processing time is required. Similarly, there is a problem in that the method of associating the approximate points with the dictionary by the DP method requires much time.

Fourth, in the case of extracting the characteristic points of the stroke, the area extraction method has a problem in that it is strong against noise but cannot extract the acute angle. On the other hand, the angle extraction method has a problem that it is easily affected by noise.

Fifth, if the characters are continuously input, information for detailed identification, for example, the number of loops and the cross position become ambiguous, which causes a problem that the detailed identification becomes difficult. Sixth, with a personal mobile terminal, there is a case where a comfortable recognition speed is desired to be provided instead of inputting in a square. On the other hand, it was not possible to set it flexibly, for example, at the expense of recognition speed, but also wanting to recognize broken letters. In order to shorten the recognition time, it is necessary to use a high-performance CPU or dynamically raise the system clock.
However, naturally, there is a trade-off relationship between the recognition speed and the life time of the battery. As described above, in the conventional system, there is a problem in man-machine property that the entire recognition speed cannot be changed according to the handwriting input state of the user, that is, the degree of breakage.

According to the present invention, the dictionary size is small, feature points can be stably extracted, and on the basis of the skeleton information of characters, on-line handwritten character recognition can be performed with high performance from regular writing to continuous input. To provide the equipment.
Another object of the present invention is to provide an on-line handwritten character recognition device capable of associating the degree of combed characters in the input pattern of the user with the recognition speed.

[0017]

In order to achieve the above object, the present invention provides a coordinate input device for inputting handwritten coordinates, a feature extraction unit for extracting feature points from the handwritten coordinate sequence, and a feature extraction unit for obtaining the feature points. In the online handwritten character recognition device including a recognition unit that recognizes a handwritten character by classifying and comparing the obtained characteristic pattern and a dictionary pattern, and a recognition result display unit that displays the recognition result, a plurality of point sequences of the coordinate point sequence From the feature amount, the feature extraction unit that extracts the feature point of the handwritten coordinates by comparing with the threshold value, the recognition dictionary unit that registered the standard character pattern including the off stroke, for each of the input pattern and the dictionary pattern, It is characterized in that it is configured by a backbone vector correspondence matching unit that sequentially performs matching processing based on corresponding points corresponding to vectors in four directions of up, down, left and right.

Further, a detailed identification section is provided for identifying a specific portion of the input character pattern by adding dictionary off-stroke information to the recognition candidate character group obtained by the backbone vector correspondence matching section. I am trying.

The feature extraction unit for extracting the feature points from the handwritten coordinate sequence is characterized by detecting the area and angle variation information consisting of a plurality of point sequences.

Further, as a classification of the recognition unit for recognizing handwritten characters by comparing the classifications, the recognition unit is characterized by performing the stroke length difference between the dictionary pattern and the input pattern.

Further, the invention is characterized in that a breaking degree setting unit is provided, and a value set by the breaking degree setting unit and a classification threshold value of the recognition unit for recognizing handwritten characters are controlled. Further, the present invention is characterized in that means for determining the degree of breakage from the feature amount of the dictionary pattern and the input pattern and controlling the system clock of the handwritten character recognition device from the degree of breakage is provided.

[0022]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will now be described with reference to FIGS.
This will be described with reference to FIG.

FIG. 1 shows the overall block configuration of an embodiment of the present invention. Reference numeral 1 is a coordinate input device TB, and 2 is a liquid crystal device LCD. The devices 1 and 2 are stacked. The coordinates of the coordinate input device TB are detected by the stroke input / output unit 4, and the handwriting is fed back and displayed according to the movement of the pen 3. The input stroke coordinate point sequence is input to the character recognition unit 5 and output as a character candidate. The output character candidate group is input to the Japanese processing unit 6, and it is determined whether or not it has meaning as Japanese by referring to the Japanese dictionary unit 7, and the result is displayed on the recognition result display unit 8. The character recognition unit 5 extracts a feature from the coordinate point sequence, the feature extraction unit 100 detects the corresponding point as a skeleton of the character in the backbone vector correspondence matching unit 200, and performs matching processing with the recognition dictionary unit 400. . Further, the detailed identification unit 300 performs a detailed identification process from the candidate character group to improve the recognition performance.

In addition, in the breakage degree setting unit 9, which is another feature, the user sets whether to input in a square pattern or continuous characters by using an icon as shown in FIG. The processing speed can be changed flexibly. In other words, the recognition speed becomes faster in the case of inputting a regular print,
In continuous character input, the recognition level is identified in detail, but it requires a high processing speed. Of course, the icon in FIG. 19 can also set intermediate values.

Next, a hardware configuration according to an embodiment of the present invention will be described with reference to FIG. In FIG. 2, the bus of the computer (CPU) 10 includes a coordinate input controller (TBC) 11, a liquid crystal controller (LCDC) 12,
Memory (MEM) 13, external communication controller (N
ET) 14 is connected. In addition, memory (MEM)
An operating system (OS), programs, and various dictionaries are stored in the storage unit 13.

The detailed flow of the character recognition unit 5, which is a characteristic part of the block configuration shown in FIG. 1, will be described below.

FIG. 3 is a detailed flow of the feature extraction unit 100 for extracting the feature points (approximate points) of the input coordinate point sequence (stroke). First, the input coordinate point sequence (eg:
In order to normalize (day), one stroke writing processing for connecting end points is also performed for off stroke (step 110). Next, resampling processing including off-stroke is performed (step 120). This process interpolates the coordinates at equal intervals, and is for eliminating the influence of the sample speed of the input coordinate point sequence and the input speed. Next, in step 130, normalization processing is performed. It is calculated as follows. Add all the lengths of the resampling coordinate point sequence obtained earlier,
The center of gravity position is calculated by dividing it by the number of coordinate point sequences. Next, the size of the input data is obtained from the maximum and minimum values of the resampled coordinate point sequence on the X and Y axes, and normalization is performed with the center of gravity position obtained earlier. In the example of FIG. 3, it can be seen that the coordinate point sequence (step 120) that is input small in the lower left is normalized and largely normalized to the center of the frame. By doing so, the influence of the input position and the input size is reduced. If the center-of-gravity position is calculated without including the off-stroke, the stroke input at a distant position has a great influence on the center-of-gravity position. Therefore, as in the embodiment of the present invention, resampling including the off-stroke is performed once, and the position of the center of gravity is obtained using this point sequence.

Next, an approximation process is carried out to obtain a feature point for recognition (step 140). This process is a process for obtaining character feature points, and is closely related to the matching process described later. The method of obtaining the feature points at equal intervals or with equal division is simple in processing, but has the drawback that the number of feature points increases and the dictionary size increases. Therefore, in one embodiment of the present invention, the minimum feature points necessary to represent the shape of a character are extracted. This is because the straight line part is only the end point,
The bent portion may be configured by a plurality of points. Therefore, there is no approximate point in the off stroke. This will be described below with reference to FIG.

FIG. 4 is an enlarged view of a part of the input coordinate point sequence, starting from the starting point Ps, Pa, Pb, P.
c, Pd ... And so on. Here, in one embodiment of the present invention, the feature point is determined by using both the area method and the angle method. Now, including the starting point Ps, the area of a triangle composed of the sequence of three coordinate points sequentially input is ΔSa, Δ
Sb, ΔSc, ..., The internal angles of the triangle are θa, θb, θc,
Then, when one of the following judgment formulas is satisfied,
Let it be a feature point.

(1) (cumulative area (ΔSa + ΔSb + ΔSc
+ ...)> threshold TH1) and (internal angle of triangle (θ
a, θb, θc, ...)> A threshold value TH2) and a feature point.

(2) Interior angles of the triangle (θa, θb, θc,
...) <The point where the threshold value becomes TH2 and the characteristic point.

As described above, the cumulative area is large, and
The condition when the interior angle is greater than the threshold works well for curve detection. Furthermore, when the interior angle of the triangle is smaller than the threshold value, this is also set as a feature point, so that the beginning of a stroke with a turn can be detected as a feature point.

FIG. 5 illustrates a specific flow of the approximation processing 140. First, the cumulative area St is cleared, the starting point Ps is read, and the coordinate buffer P
It is stored in (1) (step 141). Next coordinate point Pa,
Pb is also read and stored in the coordinate buffers P (2) and P (3) (step 142). Area ΔS from the coordinates of 3 points
To calculate the cumulative area St (step 14
3). Similarly, the internal angle θ is calculated from the coordinates of the three points (step 144). Next, the above-obtained cumulative area St and internal angle θ
Are respectively compared with thresholds TH1 and TH2 (step 1
45). If none of the conditions is satisfied, the next coordinate point Pc is read and the coordinates of the next three points are prepared (step 146). When the above condition is satisfied, the coordinate immediately before the last fetched coordinate is stored as an approximate point (step 147). Finally, it is determined whether or not the coordinates of all the sample points have ended (step 148). If not completed, the obtained approximation point is set as a new start point and the next approximation point is obtained (step 149). As described above, in the present invention, the approximate points are obtained using the area method and the angle method.

Next, the backbone vector correspondence matching process, which is a characteristic process of the present invention, will be described. At first,
Explain the concept. The character shape is basically composed of backbones in four directions (d1, d2, d3, d4) as shown in FIG. And
The diagonal direction is the auxiliary vector (d12, d23, d34, d41) of the backbone vector. Since this auxiliary vector is an intermediate vector, it can be any of the backbone vectors on both sides.

How to correspond the approximate points of the input stroke and the approximate points of the dictionary using the above-mentioned backbone vector will be described in each case of FIGS.
In each figure, the circles are the approximate points of the dictionary pattern,
The ● marks are the approximate points of the input pattern. The vectors between adjacent approximation points are a1 and a2 vectors for dictionary patterns and b1 and b2 vectors for input patterns. The corresponding points are processed as follows. (1) First, the head vectors a1 and b1 are assumed to correspond (marked with ■).

(2) Hereinafter, vector comparison, advance of the input pattern vector, and advance of the dictionary pattern vector are performed depending on the directions of the dictionary pattern and the input pattern.

(3) The above (2) is processed until there are no more approximation points.

The example of FIG. 7 will be described. First, assuming that the vectors a1 and b1 correspond first, the next a
2, b2 vectors are compared, and the same direction ((d in FIG. 6 is
When it is 1), it is moved forward to a2 and b2 as a temporary correspondence (square mark).

In the example of FIG. 8, similarly, vectors a1 and b
1 corresponds first, and next, the direction of the next a2 is judged to be b1. If they are the same vector, a2 is provisionally corresponded and it is advanced.

FIG. 9 shows an example in which the vector of the input pattern and the vector of the dictionary pattern are in opposite directions.

Similar to FIG. 7, it is assumed that the vectors a1 and b1 correspond first, and the b2 vector does not correspond because it is in the opposite direction to the dictionary vector a2. Next, the previous dictionary vector a3 is compared with the b2 vector, and if they are the same, a2
The vector is advanced so that the a3 vector temporarily corresponds to the b2 vector.

In the example of FIG. 10, the dictionary pattern vector a is used.
In this example, 1, a2 and a3 are moved forward one after another to temporarily correspond to the vector b1 of the input pattern.

As described above, it is basically important for the subsequent matching process to search for a corresponding point with the dictionary vector based on the backbone vector in four directions. That is, when the distance value is calculated in the matching process, the correspondence is made in the basic skeleton direction, and then the distance value is calculated between the corresponding points, so that the number of calculations is reduced as compared with the correspondence calculation by general DP matching. Have.

FIG. 11 shows (a) input pattern and (b) input pattern.
The dictionary pattern of "day", (c) "district" FIG. 7 to FIG.
Each case will be described. In each figure, the circle marks are the approximation points of the dictionary pattern, and the circle marks are the approximation points of the input pattern. The vectors between adjacent approximation points are a1 and a2 vectors for dictionary patterns and b1 and b2 vectors for input patterns. The corresponding points are processed as follows. (1) First, the head vectors a1 and b1 are assumed to correspond (marked with ■).

(2) Hereinafter, vector comparison, advance of the input pattern vector, and advance of the dictionary pattern vector are performed depending on the directions of the dictionary pattern and the input pattern.

(3) The above (2) is processed until there are no more approximation points.

Now, the example of FIG. 7 will be described. First, assuming that the vectors a1 and b1 correspond first, the next a
2, b2 vectors are compared, and the same direction ((d in FIG. 6 is
When it is 1), it is moved forward to a2 and b2 as a temporary correspondence (square mark).

In the example of FIG. 8, similarly, vectors a1 and b
1 corresponds first, and next, the direction of the next a2 is judged to be b1. If they are the same vector, a2 is provisionally corresponded and it is advanced.

FIG. 9 shows an example in which the vector of the input pattern and the vector of the dictionary pattern are in opposite directions.

Similar to FIG. 7, it is assumed that the vectors a1 and b1 correspond first, and the b2 vector does not correspond because it is in the opposite direction to the dictionary vector a2. Next, the previous dictionary vector a3 is compared with the b2 vector, and if they are the same, a2
The vector is advanced so that the a3 vector temporarily corresponds to the b2 vector.

In the example of FIG. 10, the dictionary pattern vector a is used.
In this example, 1, a2 and a3 are moved forward one after another to temporarily correspond to the vector b1 of the input pattern.

As described above, basically, it is important for the subsequent matching process to search for corresponding points with the dictionary vector based on the backbone vectors in four directions. That is, when the distance value is calculated in the matching process, the correspondence is made in the basic skeleton direction, and then the distance value is calculated between the corresponding points, so that the number of calculations is reduced as compared with the correspondence calculation by general DP matching. Have.

FIG. 11 shows (a) input pattern and (b) input pattern.
The relationship between the dictionary pattern of "day" and the dictionary pattern of (c) "ward" is shown. The ● marks indicate the corresponding approximate points, and the ○ marks indicate the non-corresponding approximate points. As described above, almost corresponding points can be obtained from the backbone vectors in the four directions. Here, it should be noted that when obtaining the corresponding points, the backbone vector is detected regardless of the on-stroke and the off-stroke. As a result, even if the input pattern is continuously written as shown in (a), it can be associated with the vector with the dictionary pattern for regular writing. This is effective in reducing the dictionary size and the processing time.

Next, when the corresponding points are obtained, based on this,
The distance value is calculated, and the method of calculating the vector difference, which is the basic discriminating element, will be described first. FIG. 12 shows the basic concept of vector difference calculation. Here, the ◯ mark is a dictionary pattern, and the ● mark is an input pattern.
Further, the eight-dotted line is assumed to correspond to the characteristic points.

When the number of line segments between corresponding points in case 1 is 1: 1 between the dictionary pattern and the input pattern, each vector a
The difference between 1 and b1 is calculated.

When the number of line segments between corresponding points in case 2 is 1: n, the other line segment is divided according to the ratio of the lengths of the n line segments, and the accumulation of the respective vector differences is calculated. This often occurs when one line segment is one, for example, in the case of a straight stroke such as "1" or "one".

As in case 3, the line segment between corresponding points is n:
In general, most cases are n. In this case, a vector of a long line segment in the corresponding line segments is divided at a ratio of the line segments in between, and the vector difference is calculated.
For example, a12 and b2 calculate the vector difference like a21 and b3. Then, when the difference in the lengths of the line segments is small, the difference is calculated between the vectors (eg, a11).
And b1). Finally, there is no corresponding line segment, and the length of the remaining line segment is calculated as a penalty value.

As described above, when there is an uncorresponding feature point (approximate point) centered on the corresponding point, the virtual corresponding point is calculated while dividing the line segment, and the accuracy of the distance value calculation is calculated. It is a feature of the present invention that the improvement and the simplification of the calculation are achieved. The calculation of the vector difference is performed from the start point to the end point of the stroke, but when the vector difference is larger than a predetermined value, it is determined that the character is not a candidate and the distance value calculation is cut off to speed up. Can be implemented.

Although the method of calculating the vector difference as the basic discriminating element for the distance value calculation has been described above, the distance value is actually calculated in consideration of other discriminating elements. FIG. 13 describes the entire recognition distance value calculation. In addition to the vector difference V (ab) described above, the coordinate difference P (ab) between corresponding points and the start / end point difference Pse (ab- between stroke start and end points)
b) and finally, there is a stroke length difference L (ab). In addition, the recognition distance value Φ includes coefficients α, β, and
It is assumed that the weighted average is obtained by multiplying γ and δ. Here, the meaning of each discriminating element is the vector difference V (ab)
Is an element for grasping the skeletal state of the entire character, and the coordinate difference P (a
-B) and the start / end point difference Pse (ab) are elements for grasping the positional relationship between corresponding points, and the final stroke length difference L (ab).
Is an element that represents the complexity of a character.

The processing flow of the above-described backbone vector correspondence matching method will be described with reference to FIGS. 14 to 16. FIG. 14 is a main processing flow of the backbone vector correspondence matching unit 200. Step 210
~ 230, the corresponding points are processed and the matching process is performed after the main classification. That is, first,
Collection information such as input stroke length is collected (step 210). From this information and the information stored in the dictionary,
Excludes characters that are not needed for matching. In one embodiment of the present invention, the following processing is performed only when the input stroke count + 2 is smaller than the dictionary stroke count (step 220). This is because the dictionary stores patterns in regular writing, but it is judged that the number of input strokes is smaller than that in this dictionary. In other words, to be precise, this occurs because the portion to be subsequently input is divided or written, or an unnecessary number of strokes is input. Generally, no more than two strokes are input than the number of strokes in the dictionary. Then another 1
As one classification, matching is performed only when the difference between the input stroke length and the dictionary stroke length is smaller than the threshold value TH3 (step 230). This processing omits matching with a character that greatly differs from the input stroke length, and is necessary for speeding up the processing. After the above classification is completed, the above-mentioned backbone vector correspondence processing is performed (step 240). The detailed flow will be described with reference to FIG. Then, finally, a matching process, which is a distance value calculation, is performed based on the obtained correspondence vector (step 250).
The above process is performed for all dictionaries (step 26).
0).

FIG. 15 is a detailed flow of the backbone vector correspondence processing (FIG. 14, step 240). Figure 7
~ Processing is performed according to the algorithm described above in Fig. 10. First, the approximate point at the beginning of the stroke is set as the corresponding point (step 240a). Next, the dictionary point and the input point are respectively advanced (step 240b). The vector is calculated, and it is determined whether they are in the same direction (step 240c). Whether or not the directions are the same is determined by the backbone vector of FIG. The intermediate auxiliary vector is determined to be a vector on both sides as described above. If it is determined in step 240c that they are in the same direction, the process jumps to step 240d to advance the dictionary point and the input point to the approximate point. On the other hand, in step 240c, when the direction is different, step 24
Jump to 0 g, move forward the dictionary point, and similarly determine the direction. If they are in the same direction, dictionary points are set as corresponding point candidates (steps 240g to 240I). Next, at the input point as well, it is similarly advanced to determine the direction (steps 240j to 240l). Now, in step 240c, if the next vector is also in the same direction when it is in the same direction (step 240e), if it is in the same direction, both dictionary points and input points are set as corresponding point candidates (step 24).
0f). If the direction is different, the process jumps to step 240g. The above process is performed for all coordinate points (step m).

FIG. 16 is a processing flow for calculating the recognition distance value based on the correspondence vector obtained by the backbone vector correspondence processing. This processing is to obtain each discriminant element as described in FIG. 13 and weight the average of the coefficients. First, the element of the vector difference V (ab) is calculated (step 250a). As described in FIG. 12, this is to cumulatively add the differences between the dictionary vector and the input vector at all corresponding points. Next, the identification element of the coordinate difference P (ab) is calculated (step 250b), and the start / end point difference Pse is calculated.
The calculation of the identification element of (ab) (step 250c) and the calculation of the identification element of the stroke length difference L (ab) (step 250d) are performed. Finally, the recognition distance value Φ is obtained by multiplying each discriminant element by a coefficient and performing a weighted average (step 25).
0e). The above process is performed for all the classified characters (step 250f). Finally, the candidate character group can be obtained by sorting the obtained distance values in ascending order. Usually, the 10th place is the candidate,
After that, it is rejected.

Next, the detailed identification unit 300 of FIG. 1 makes a detailed determination on the obtained candidate character group by the backbone vector correspondence matching method. First, the concept of detailed identification will be described, and then the processing flow will be described.

FIG. 17 describes the basic method of detailed identification. Each detailed identification element is classified into a common processing for characters and a processing for each character. Note that the common processing reflects the recognition distance value as a penalty, but the character-by-character processing replaces the candidate ranks. First, shape comparison is performed. As shown in the figure, the shape is detected by dividing it into seven segments of straight line, right loop, right curve, right corner, left loop, left curve, and left corner.
Since off-stroke information is added to the dictionary, it is possible to determine the shape of the specific portion by temporarily allocating the off-stroke information to the input pattern using this information. For example, "ha" has a right loop,
"Ke" can be identified as having a right curve. As common processing, for example, detailed identification is performed by adding a difference in the number of loops as a penalty. The character-by-character processing, in particular, creates a table of similar character groups and compares only those characters.

The next detailed identification element is to compare relative positions between strokes, and as shown in FIG.
The presence / absence of four types of cross type, T type, L type, and Y type is compared. For example, "stone" is detected as having no cross, and "right" is detected as having cross. It should be noted that this relative position comparison is performed by obtaining the crossover information from the straight lines of the related vectors. This relative position comparison also performs both common processing and character-by-character processing.

In the above-mentioned backbone vector correspondence matching method, basically, the stroke number information is not used and it is processed as one stroke. Therefore, the same recognition algorithm can be used to recognize from regular characters to consecutive characters. However, on the contrary, this causes a group of characters that have almost the same shape, which reduces recognition performance. For example,
"Otsu" and "2" are identified as the same pattern. Therefore, the number of strokes is compared for each character. The correct number of strokes is stored in the dictionary, so if the input is 2 strokes, change it to "2".
If it is one stroke, it will be called "Oto". Similarly, there are some characters that can be identified for the stroke length as well (eg "end" and "end", "!" And ":"). This can be identified by comparing the lengths of the horizontal and vertical bars.

As the last element of the detailed identification, radicals are compared for each character. In the present invention, even if there is an off stroke,
Since it is a method of recognizing as one stroke, there are characters that are not clearly distinguished. Therefore, only the radicals that differ are cut out and compared (eg: “timber”, “juice”, “hiro”).
And "pay").

FIG. 18 shows the processing flow of the above detailed identification algorithm. In FIG. 18, first, a loop between the input pattern and the dictionary pattern is detected, and if there is an input and it is not in the dictionary, a penalty is added to the character (step 310). Also, the connection information between the input pattern and the dictionary pattern is detected, and a penalty is added depending on the presence / absence of a cross, for example (step 320). Since these two processes are common processes, based on the value with the penalty added,
It needs to be re-sorted (step 330). next,
It is determined whether or not the first ranked character of the sorted candidate is a character tabulated for each character (step 340).
If the character defined in the table is in the first place, the above detailed identification processing is performed (step 350). If the first-ranked candidate character is in the table, corresponding characters of other candidates also exist, and therefore, the character including that character is compared in more detail. If a character other than the correct character is closer to the input pattern, the candidates are replaced.

As described above, since the dictionary pattern is registered in a regular form, the off stroke information is used to provisionally assign the off stroke to the input pattern as well, whereby the details of the character can be identified and the recognition rate can be improved. Can be improved.

Next, details of the breakage degree setting unit 9 shown in FIG. 1 will be described with reference to FIGS. 19 to 21. First,
FIG. 19 shows a breakage degree setting panel. This panel shows the status of (a) regular input, (c) continuous character input, or (b) intermediate input. When the input is set to the regular writing, as shown in FIG. 20, the threshold value TH3 of the stroke length difference between the input stroke and the dictionary stroke is reduced. On the contrary, if you use continuous letters,
Since the stroke is deformed a lot, the difference in stroke length also varies greatly. Therefore, it is necessary to increase the threshold value TH3. This means that the classification width in FIG. 14 is changed, and is closely related to the recognition time. Therefore, the user can enjoy a comfortable recognition speed if he / she inputs it in a regular form, but the recognition rate of the broken character becomes worse. on the other hand,
If you have a lot of comb letters, put them on the side of the letters
The recognition rate is improved, but the recognition speed is sacrificed. As described above, by changing the threshold value of character classification, it is possible to deal with various input forms, and it is particularly effective for mobile terminals.

FIG. 21 illustrates the processing flow of the crushing degree setting unit 9. This can be realized simply by reading the value on the setting panel for the degree of breakage (step 14a), then drawing and setting the threshold value table according to this value (step 14b). Although the threshold for classification is used as the threshold for setting the degree of breakage, it may be a threshold for other matching processing. On the contrary, the difference between the number of strokes of the recognized character and the number of strokes in the dictionary is detected, and if the difference in the number of strokes is small, it is determined whether or not the input is a cross-script input. By setting the clock late, the battery life of the mobile terminal can be extended. Of course, in the opposite case, when the difference in the number of strokes is large, it is determined that the input pattern has many continuous characters, and the clock of the CPU is accelerated. At this time, the battery will be consumed faster, and the battery life will be shortened. In any case, by controlling the width of classification, it is possible to control battery consumption and control the recognition processing speed.

[0072]

As described above, according to the embodiments of the present invention, the following effects can be obtained. First, by dividing the skeleton of the character into backbone vectors with a certain angle and then performing matching processing, flexible input from regular writing to continuous writing can be recognized with only one dictionary pattern. It is possible. This eliminates the need for registering a plurality of dictionaries such as Kushiji or Kanji, which has the effect of reducing the dictionary size. Further, by performing the backbone vector correspondence processing and the matching processing unit separately, the correspondence of the approximate points becomes simple, and the processing time of the matching can be reduced.

As a second effect, in the case of extracting the characteristic points of the coordinate point sequence, the area method for comparing the area of the triangle formed by the plurality of coordinate point sequences with the threshold value and the plurality of coordinate point sequences are used. By combining the angle method of comparing the interior angle with the threshold value, it is possible to stably obtain a small number of approximation points. This has the effect of reducing the size of the dictionary.

As a third effect, in the backbone vector corresponding matching method, even if an uncorresponding approximate point is tentatively corresponded from the peripheral corresponding points, there is an effect that the distance calculation of the matching can be accurately performed. Therefore, the recognition rate is also improved.

As a fourth effect, by providing the breakage degree setting, the threshold value can be dynamically changed and the recognition processing time can be changed. As a result, a comfortable recognition speed can be provided to a user who does not input a lot of kanji characters, such as a regular writing method.

As the fifth effect, the power consumption can be controlled and the life of the portable terminal can be extended by determining the degree of breakage based on the difference in the number of strokes from the dictionary and linking it with the clock speed of the processing system. it can.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a block configuration of an embodiment of the present invention.

FIG. 2 is a diagram illustrating a hardware configuration of an embodiment of the present invention.

FIG. 3 is an overall processing flow of a feature extraction unit according to an embodiment of the present invention.

FIG. 4 is a diagram illustrating an algorithm of a feature extraction unit according to an embodiment of the present invention.

FIG. 5 is a feature point extraction processing flow of a feature extraction unit according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating a backbone vector according to an embodiment of the present invention.

FIG. 7 is a diagram illustrating correspondence between backbone vectors according to an embodiment of the present invention.

FIG. 8 is a diagram for explaining backbone vector correspondence according to an embodiment of the present invention.

FIG. 9 is a diagram illustrating correspondence between backbone vectors according to an embodiment of the present invention.

FIG. 10 is a diagram for explaining backbone vector correspondence according to an embodiment of the present invention.

FIG. 11 is a diagram illustrating correspondence between backbone vectors according to the embodiment of this invention.

FIG. 12 is a diagram illustrating a vector difference calculation algorithm according to an embodiment of the present invention.

FIG. 13 is a diagram illustrating a recognition distance value calculation algorithm according to an embodiment of the present invention.

FIG. 14 is a matching processing flow according to an embodiment of the present invention.

FIG. 15 is a backbone vector correspondence processing flow according to an embodiment of the present invention.

FIG. 16 is a recognition distance value calculation processing flow according to an embodiment of the present invention.

FIG. 17 is a diagram illustrating an algorithm for detailed identification according to an embodiment of the present invention.

FIG. 18 is a processing flow of detailed identification according to an embodiment of the present invention.

FIG. 19 is a diagram illustrating a scraping degree setting panel according to an embodiment of the present invention.

FIG. 20 is a diagram for explaining the relationship between the breakage level setting and the threshold value according to the embodiment of the present invention.

FIG. 21 is a processing flow for setting a scrap level according to an embodiment of the present invention.

[Explanation of symbols]

1 ... Coordinate input device, 2 ... Liquid crystal device, 3 ... Pen, 4 ... Stroke input / output unit, 5 ... Character recognition unit, 9 ... Breakage level setting unit, 200 ... Backbone vector matching unit,
300 ... Detail identification part, 400 ... Recognition dictionary part.

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masayuki Ishida 1-16-5 Dogenzaka, Shibuya-ku, Tokyo Inside Nikkei Information Systems (72) Inventor Kayoko Taki 1-16-5 Dogenzaka, Shibuya-ku, Tokyo In Information Systems (72) Inventor Ryunosuke Murao 1-16-5 Dogenzaka, Shibuya-ku, Tokyo Inside Information Systems (56) Reference JP-A-5-225396 (JP, A) JP-A-8-180135 (JP , A) JP 9-16720 (JP, A) JP 8-101889 (JP, A) JP 3-282896 (JP, A) JP 60-112187 (JP, A) JP 8-305851 (JP, A) JP-A-7-129720 (JP, A) JP-A-7-73275 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06K 9/00 -9/82

Claims (5)

(57) [Claims] 1. A coordinate input device for inputting handwritten coordinates, a recognition unit for recognizing handwritten characters by classifying and comparing a feature pattern obtained from a coordinate point sequence of the handwriting coordinates, and a dictionary pattern, and the recognition unit. In an online handwritten character recognition device having a recognition result display unit that displays a recognition result, the recognition unit obtains a feature amount from a plurality of point sequences of the coordinate point sequence, and compares the handwriting coordinates with a threshold value. The feature extraction unit that extracts the feature points, the recognition dictionary unit that registers the standard character patterns including off-strokes, and the corresponding points that correspond to the four vertical and horizontal vectors for each of the input pattern and the dictionary pattern. Originally, it is obtained by the backbone vector corresponding matching unit that sequentially performs matching processing and the backbone vector corresponding matching unit .
Corresponding vector corresponds to the backbone vector
After that, the recognition distance value is calculated, and the recognition candidate is calculated from the recognition distance value.
The character group is obtained, and the dictionary
Input part of character pattern by adding stroke information
An online character recognition device having a detailed identification section for identifying minutes . 2. A coordinate input device for inputting handwritten coordinates, and a characteristic pattern obtained from a coordinate point sequence of the handwritten coordinates.
And recognize handwritten characters by classifying and comparing dictionary patterns
A recognition unit and a recognition result display unit that displays the result recognized by the recognition unit.
In the on-line handwritten character recognition device having the above, the recognizing unit accumulates an area from a plurality of coordinate point sequences.
The product value and the angle are feature values, and the cumulative value of the area is the threshold value.
Is greater than Th1 and the angle is greater than the threshold Th2.
Larger or the angle is smaller than the threshold Th2
Sometimes, the point before the last captured coordinate is used as an approximation point.
A feature extraction unit that extracts feature points of handwritten coordinates and recognition that registers standard character patterns including off-strokes
For the dictionary part, input pattern and dictionary pattern,
Based on the corresponding points corresponding to the vectors in the four directions,
Backbone vector matching unit that performs ching processing
Online character recognition apparatus characterized by having, when. 3. A coordinate input device for inputting handwritten coordinates, and a characteristic pattern obtained from a coordinate point sequence of the handwritten coordinates.
And recognize handwritten characters by classifying and comparing dictionary patterns
A recognition unit and a recognition result display unit that displays the result recognized by the recognition unit.
In the on-line handwritten character recognition device having, the recognizing unit calculates a feature amount from a plurality of point sequences of the coordinate point sequence.
And extract the feature points of handwritten coordinates by comparing with the threshold value
Feature extraction unit and recognition that registers standard character patterns including off-strokes
For the dictionary part, input pattern and dictionary pattern,
Based on the corresponding points corresponding to the vectors in the four directions,
Backbone vector matching unit that performs ching processing
And a classification of a recognition unit for recognizing handwritten characters by comparing the classifications with
Then, the recognition unit obtains the stroke lengths of the input pattern and the dictionary pattern only when the input stroke count + 2 is smaller than the dictionary stroke stroke,
Stroke length of force pattern and stroke of dictionary pattern
Matches with a dictionary pattern when the difference in length is less than a threshold
If the difference exceeds the threshold, the dictionary pattern and
An online character recognition device that is not touched . 4. A coordinate input device for inputting handwritten coordinates, a feature extraction unit for extracting feature points from the handwritten coordinate sequence, a feature pattern obtained from the feature extraction unit, and a dictionary pattern for classification and handwriting. In an online handwritten character recognition device having a recognition unit for recognizing characters and a recognition result display unit for displaying the recognition result, the degree of breakage is obtained from the feature amount of the dictionary pattern and the input pattern, and the handwriting is performed based on this degree of breakage. An online character recognition device comprising means for controlling a system clock of the character recognition device. 5. The method of claim 4, as a feature of the breaking of the online character recognition device which is a number of strokes difference dictionary pattern and the input pattern.