JPH11203408A - Handwritten pattern storing/retrieving device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a handwritten pattern storing/retrieving device capable of executing highly accurate retrieval by absorbing the unstability of division into partial patterns due to maintenance of a high speed property. SOLUTION: The device includes a partial pattern separation means 103 for separating a handwritten pattern into partial patterns and a feature extraction means 106 for extracting a feature from each separated partial pattern and a feature matching means 112 matches the features of stored partial patterns with the feature of a retrieving key pattern and reads out a handwritten pattern corresponding to the matched partial pattern. At the time of matching the features of partial patterns by the means 112, plural partial patterns are synthesized, the feature of the synthetic pattern is found out and features are matched by using the features of the original partial patterns or the feature of the synthetic pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten pattern storage and retrieval apparatus for retrieving a stored handwritten pattern, and more particularly to a handwritten pattern storage and retrieval apparatus for retrieving a stored pattern similar to a handwritten pattern inputted at the time of retrieval. About the method.

[0002]

2. Description of the Related Art Conventionally, this type of handwritten pattern storage and retrieval apparatus stores a handwritten pattern and, as shown in, for example, JP-A-8-305722, stores the handwritten pattern in a search key pattern. Is used for searching and outputting a storage pattern similar to the key search pattern. Hereinafter, the above-described handwritten pattern storage and retrieval device is referred to as a first related art.

FIG. 13 is a block diagram showing an embodiment of a conventional handwritten pattern storage and retrieval apparatus. A handwriting pattern input unit 101 samples a pattern handwritten using a pen with a tablet device or the like by using two-dimensional coordinates of a movement of a pen tip while the pattern is being written in time series. Output the value. Here, a coordinate system is used in which the origin is located at the upper left corner of the tablet, the X coordinate increases as moving to the right, and the Y coordinate increases as moving down. The handwritten pattern buffer 102 is a temporary storage unit for handwritten patterns. The character separating unit 2030 separates the handwritten pattern for each character. Character code and pointer storage means 209
0 records the character code of the recognition result, the reliability of the recognition result, and the position in the original handwritten pattern corresponding to the character. The handwritten pattern storage unit 108 stores the handwritten pattern input from the handwritten pattern input unit 101. The character code and pointer buffer 2110 stores a character code as a result of recognizing a handwritten pattern input as a search key, the reliability of the recognition result, and a pointer indicating the position of the character in the original handwritten pattern. . The character code reading unit 2100 sequentially reads the character code and the reliability of the recognition result from the character code and pointer storage unit 2090, and reads the character code matching unit 2120.
Transfer to The character code matching means 2120 detects a matching part using the character code reading means 2100, the character code transferred from the character code and pointer buffer 2110, and the reliability of the recognition result. The handwritten pattern display means 114 displays the handwritten pattern of the search result.

Next, the operation of the above-described handwritten pattern storage and retrieval apparatus will be described. Handwriting pattern input means 10
1 and the handwritten pattern output from it is
For example, it is output in a format as shown in FIG. The coordinates of the point indicated by the moving pen are sampled at regular time intervals, and the data is output as a vertical column in FIG. One sampled data includes the serial number of the point, the number of the page on which the handwritten data is written, the X and Y coordinates in the page, and whether the pen tip is in contact with the tablet or not. Up / do
wn signal. When the pen tip is in contact with the tablet, 1 as an up / down signal, that is, down
A signal is output, and when the pen tip is not in contact with the tablet, 0, that is, an up signal is output.

[0005] The handwritten pattern input from the handwritten pattern input means 101 is stored in a handwritten pattern buffer 1.
02 is temporarily stored. Character separation means 2030
The handwritten pattern received from the handwritten pattern buffer 102 is cut for each character. In this conventional example, a plurality of frames are displayed on the handwritten pattern input means 101, and one character is entered in one frame. Therefore, the upper left coordinates (X1i, Y1i) of the predetermined frame are displayed. Utilizing the lower right coordinates (X2i, Y2i) of the frame, it is detected that a character is written in the i-th frame, and as long as it is written continuously in one frame, it is input. Record the dots as one character.

In order to determine whether or not a point entered in the i-th frame is present, the coordinates of the point are determined by (x, y).
When x is greater than or equal to X1i, x is less than or equal to X2i, and y is greater than or equal to Y1i, and y is less than or equal to Y2i, it is assumed that the information is entered in one frame. And If a point that does not satisfy the frame determination condition is input, it is determined that the point immediately before that point has been entered in the i-th frame, and that point is separated as one character and the character Transfer to 2060. For the points input thereafter, it is searched whether another frame satisfies the frame determination condition. Then, a frame having an upper left coordinate and a lower right coordinate satisfying the frame determination condition is selected, and it is checked whether or not a point input as the i-th frame thereafter satisfies the same frame determination condition.

In the same manner as described above, a point is recorded as one character as long as continuously input points continue to satisfy the same frame determination condition, and when the point is no longer satisfied, a set of previous points is set. As one character, character recognition means 2060
Transfer to The character pattern generated by cutting by the character separating means 2030 is recognized by the character recognizing means 2060. The character recognition means 2060 may be any means for recognizing a handwritten pattern input as online data. The character recognizing unit 2060 outputs a character code of the recognition result and a certainty factor of the recognition result as an output result. The certainty factor indicates that a larger value indicates a more accurate recognition result. However, when one character pattern is recognized, the character recognition means 2060 outputs a set of a plurality of character codes and certainty factors.

The recording / retrieval control means 115 connects the first switch 107 to the character code and pointer storage means 2090 and connects the second switch 105 to the handwritten pattern storage means 108 for the degree of accumulation of the handwritten pattern. .
Thereby, at the time of accumulation, the handwritten pattern input means 101
The handwritten pattern inputted from the second switch 10
5 is stored in the handwritten pattern storage means 108,
The character code generated by the character recognizing means 2060, the certainty factor, and a pointer indicating where the character to be recognized corresponds to the original handwritten pattern in which the character to be recognized corresponds are transmitted through the first switch 107 to the character code and the character code. It is stored in the pointer storage means 2090. FIG. 16 shows the format of the character code and the data stored in the pointer storage means 2090. The data shown in FIG. 16 is an example of a case where the first candidate and its certainty, and the second candidate and its certainty are stored as recognition results after performing recognition on one character. The same applies when storing more candidates. The number of the first point and the number of the last point of the point data forming one character pattern are stored as pointers to the handwritten pattern.

At the time of retrieval, the first switch 107 is connected to the character code and pointer buffer 2110 according to a signal from the record retrieval control means 115, and the second switch 105 is disconnected. A handwritten pattern serving as a key for search is input from the handwritten pattern input means 101 and temporarily stored in the handwritten pattern buffer 102. The handwritten pattern is transferred to the character separating unit 2030 and separated for each character. Data as shown in FIG. 15 is extracted and transferred to the character recognizing unit 2060 together with the original handwritten pattern. In the character recognition means 2060, 1
The result of recognition for each character is recorded in the form shown in FIG. 16, transferred to the character code and pointer buffer 2110 and temporarily stored. Next, the character code and the character data in the pointer buffer 2110 are transferred to the character code matching means 2120 one character at a time. When the character code matching unit 2120 receives the data of the first character, the character code matching unit 2120 reads the data related to the first character in the character code and pointer storage unit 2090 from the character code reading unit 2100. The one-character data read from the character code and pointer buffer 2110 includes two character codes in the example of FIG. Similarly, the character code and 1 read from the pointer storage means 2090
The character data also includes two character codes. Therefore, there are four possible combinations. From the four sets, a set in which the two character codes match is selected. If there is a matching character code,
Character code and pointer storage means 2090 and character code and pointer buffer 211
Read from 0 and calculate the confidence of the set by sum or product. In the example, calculation is performed using the product. If the character codes match in a plurality of sets, the certainty factor of the larger set is selected and stored as the total certainty factor. If the accumulated certainty satisfies a certain condition, for example, if it is larger than a predetermined threshold, the certainty is read from the character code and pointer storage means 2090 to read the next character data, and the character code and The next character data is read from the pointer buffer 2110, and the same combination confidence is calculated as described above. The obtained confidence sum is multiplied by the confidence sum to obtain a new confidence sum. After this process is performed on all the characters included in the character code and pointer buffer 2110, if the above-described cumulative certainty factor is equal to or greater than the threshold value, the character code and pointer storage means 209
A corresponding character string in 0 is selected as a search result. If there is no matching code included in the set while matching a series of character data in the character code and pointer buffer 2110, or a certainty accumulates during the matching of a series of character data. If the value is smaller than the value, or if all of the series of character strings have been matched, the matching process is stopped, and the character code and the character code that match the first character data in the character code and pointer buffer 2110 are set. The character data next to the character data in the pointer storage unit 2090 is read, the matching process with the character code and the first character data in the pointer buffer 2110 is started, and a series of character strings are matched. If a corresponding character string can be selected as a search result from the character code and pointer storage means 2090, data relating to the character is transferred to the handwritten pattern reading means 113. The handwritten pattern reading means 113 acquires the start point number of the first character and the end point number of the last character of the matched character string in the character code and pointer storage means 2090, and stores the point data from the start point to the end point in the handwritten pattern storage. Read from the means 108, and the handwritten pattern display means 1
Transfer to 14.

As another conventional handwritten pattern storage and retrieval device (second prior art), there is known one described in Japanese Patent Application Laid-Open No. 8-77295. Hereinafter, the handwritten pattern storage and retrieval device will be described. The handwritten pattern input from the online handwritten pattern input means is stored and stored for each stroke, and the handwritten pattern input as a key at the time of retrieval is also described as a stroke set, and the stroke set of the key handwritten pattern is written. Is selected as a subset from the stored stroke sets. When a similar subset of strokes can be detected from the stored data, the detection result is presented to the handwritten pattern display device.

[0011]

In the above-mentioned first prior art, the characters in the handwritten pattern are stored one by one in a frame presented beforehand, regardless of whether the pattern is an accumulated pattern or a search key pattern. Since the entered information is used as an input, it can be executed without ambiguity when separating it into one character.
However, when a handwritten pattern entered without a frame is separated for each character, the result may be different due to a slight change in the shape of the entered character. For example, "word"
Also, when writing the word "", there is a case where "word" is detected as one character, and a case where the biased "word" and the adjacent "go" are separated into two characters. Thus, "words" in the accumulation pattern
Is recorded, and the word "go" is separated from the word "go" by the search key pattern, the first conventional method cannot search well.

For example, when a pattern similar to FIG. 10 is input, it is divided into a partial pattern 1401 and a partial pattern 1402 as shown in FIG. 11, and a partial pattern 1501 and a partial pattern 1502 as shown in FIG. It can be assumed that they are separated. However, even if the separated and recognized results are accumulated as shown in FIG. 11 and the search is performed using the recognized and separated search key patterns as shown in FIG. 12, a desirable search result is not obtained. I can't.

Next, another problem in the first prior art will be described. Even if the above-described character detection is successful and the corresponding character can be separated by the storage pattern and the search key pattern, the same code may not be obtained as a result of character recognition. In particular, in the case of handwriting, the recognition result is unstable, and even if a certain number of candidates have been selected in advance, the recognition result of characters written in the same category at different times is calculated by finding the intersection set of the candidate set. May be. In such a case, the first
In the related art, desired search results cannot be obtained. The second prior art has been proposed to overcome the problems of the first prior art. However, in the second prior art, a stroke set included in a search key pattern for each stroke is set. Since a set of strokes of an accumulation pattern matching with the above is required, there is a problem that a very long calculation time is required.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a handwritten pattern storage and retrieval apparatus which has a small number of missed searches for a handwritten character pattern written in a high-speed and frameless state.

It is another object of the present invention to provide a handwritten pattern storage and retrieval apparatus capable of storing and retrieving not only handwritten characters but also handwritten patterns such as handwritten line figures using the same device as the handwritten characters.

[0016]

According to the present invention, there is provided a handwritten pattern input means and a handwritten pattern storage means for searching for a stored pattern similar to a search key pattern inputted from the handwritten pattern input means. In the handwritten pattern storage and retrieval apparatus, a partial pattern separating unit, a partial pattern feature extracting unit, a feature matching unit that matches the features of the stored partial patterns and the features of the search key pattern, and corresponds to the matched partial patterns A handwritten pattern storage and retrieval apparatus characterized by comprising means for reading a handwritten pattern to be read.

Further, according to the present invention, the feature matching unit combines a plurality of partial patterns to obtain a feature of the combined pattern, and uses a feature of the original partial pattern or a feature of the combined pattern. Thus, there is provided a handwritten pattern storage and retrieval apparatus characterized in that the apparatus is provided with matching executing means for performing feature matching.

Further, according to the present invention, the feature synthesizing means includes means for obtaining a sum of partial pattern data from characteristics of a plurality of partial patterns and generating a characteristic of the synthetic pattern as a synthetic characteristic value. Is obtained.

Further, according to the present invention, the plurality of partial pattern data includes direction counting data, starting point direction data,
A handwritten pattern storage and retrieval apparatus characterized by comprising end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data is obtained.

Further, according to the present invention, there is provided a handwritten pattern storage and retrieval device having a handwritten pattern input device and a handwritten pattern storage device for retrieving a stored pattern similar to the retrieval key pattern inputted from the handwritten pattern input device. In the apparatus, a partial pattern separating unit, a partial pattern feature extracting unit, a feature matching unit that matches a feature of the stored partial pattern and a feature of the search key pattern, and a handwritten pattern corresponding to the matched partial pattern A handwritten pattern storage and retrieval apparatus is provided, comprising a reading means and a handwritten pattern matching means for matching the read handwritten pattern with a search key pattern.

Further, according to the present invention, the feature matching means combines a plurality of partial patterns to obtain the features of the combined pattern, and uses the features of the original partial pattern or the features of the combined pattern. Thus, there is provided a handwritten pattern storage and retrieval apparatus characterized in that the apparatus is provided with matching executing means for performing feature matching.

Further, according to the present invention, the feature synthesizing means includes means for obtaining the sum of the partial pattern data from the features of the plurality of partial patterns and generating the feature of the synthesized pattern as a synthesized feature value. Is obtained.

Further, according to the present invention, the plurality of partial pattern data includes direction counting data, starting point direction data,
A handwritten pattern storage and retrieval apparatus characterized by comprising end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data is obtained.

Further, according to the present invention, there is provided a handwritten pattern storage means for retrieving a stored pattern similar to the search key pattern inputted from the handwritten pattern input means, comprising handwritten pattern input means and handwritten pattern storage means. In the apparatus, a partial pattern separating unit, a partial pattern feature extracting unit, a feature matching unit that matches a feature of the stored partial pattern and a feature of the search key pattern, and a handwritten pattern corresponding to the matched partial pattern Reading means, recognition means for executing recognition by inputting a result obtained by separating the read handwritten pattern into partial patterns by the partial pattern separating means, and converting the search key pattern into a partial pattern by the partial pattern separating means. Recognition means for performing recognition with the separated result as input, and Handwritten pattern storage retrieval device is obtained, characterized in that the results recognized handwritten pattern and is configured by including a character comparing means for comparing the result of recognizing the search key pattern.

Further, according to the present invention, the feature matching unit combines a plurality of partial patterns to obtain a feature of the combined pattern, and uses a feature of the original partial pattern or a feature of the combined pattern. Thus, there is provided a handwritten pattern storage and retrieval apparatus characterized in that the apparatus is provided with matching executing means for performing feature matching.

Further, according to the present invention, the feature synthesizing means includes means for obtaining a sum of partial pattern data from characteristics of a plurality of partial patterns and generating a characteristic of the synthetic pattern as a synthetic characteristic value. Is obtained.

Further, according to the present invention, the plurality of partial pattern data includes direction counting data, starting point direction data,
A handwritten pattern storage and retrieval apparatus characterized by comprising end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data is obtained.

[0028]

According to the present invention, since matching is checked after being separated into partial patterns in advance, it is possible to quickly search and display a portion of an accumulation pattern similar to a search key pattern.

Also, when a character is separated into two at the time of integration from a stroke into a partial pattern, or is integrated with another character, and a partial pattern that is unbalanced with the pattern at the time of accumulation is generated. Also, since multiple partial patterns are integrated and matched after obtaining the corresponding partial patterns, the pattern entered at the time of accumulation and the pattern entered at the time of retrieval are separated by character spacing and radical spacing etc. Even if there is a mismatch, it can be correctly detected.

In addition, since a method of calculating the feature of a pattern obtained by combining a plurality of partial patterns by using the sum of the features of the combined partial patterns and the selection of the maximum value and the minimum value is provided, the method can be performed at high speed. Can perform searches.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a first embodiment of the present invention.
It is a block diagram which shows the structure of the handwritten pattern accumulation search device which concerns on embodiment. In FIG. 1, a handwritten pattern input unit 101 samples a handwritten pattern using a pen with a tablet device or the like by using two-dimensional coordinates of a movement of a pen tip while the pattern is being written in time series. And outputs the sampling value. The handwritten pattern buffer 102 is a temporary storage unit for handwritten patterns. The partial pattern separating unit 103 separates a handwritten pattern into partial patterns. The feature and pointer storage unit 109 records the feature of the partial pattern and the position in the original handwritten pattern corresponding to the partial pattern. The handwritten pattern storage unit 108 stores the handwritten pattern input from the handwritten pattern input unit 101. The feature and pointer buffer 111 stores the feature of a partial pattern in a handwritten pattern input as a search key and a pointer indicating the position of the partial pattern in the original handwritten pattern. The feature reading unit 110 sequentially reads the features from the feature and pointer storage unit 109 and transfers the features to the feature matching unit 112. The feature matching unit 112 detects a partial pattern matching with the feature reading unit 110 using the feature and the feature transferred from the pointer buffer 111. The handwritten pattern display means 114 displays the handwritten pattern of the search result.

Next, the operation will be described. The handwritten pattern input to the handwritten pattern input means 101 and output therefrom is output, for example, in a format as shown in FIG. This will be described with reference to FIG. The coordinates of the movement of the pen are sampled at regular time intervals, and the data is output as a vertical column in FIG. One sampled data indicates the serial number of the point, the number of the page on which the data is described, the X and Y coordinates in the page, and whether the pen tip is touching or not touching the tablet. It consists of a pen up / down signal. When the pen tip is in contact with the tablet, 1 is output as a pen up / down signal, that is, a pen down signal is output. When the pen tip is not in contact with the tablet, 0, that is, a pen up signal is output. .

The handwritten pattern input from the handwritten pattern input means 101 is stored in the handwritten pattern buffer 1.
02 is temporarily stored. Partial pattern separation means 10
3 cuts the handwritten pattern received from the handwritten pattern buffer 102 into partial patterns.

FIG. 2 is a block diagram showing the detailed processing of the partial pattern separating means 103. The partial pattern separating unit 103 includes a point data reading unit 701 for sequentially reading point data from the handwritten pattern buffer 102, and a pen u for the point data received from the point data reading unit 701.
a stroke generating means 702 for examining the p / down signal and recording data relating to the stroke; an inter-stroke time calculating means 703 for calculating a time interval and a moving amount from the immediately preceding stroke for each stroke; Means 704, separation determining means 705 for determining whether to separate into blocks different from the previous stroke based on the time interval and the movement amount from the immediately preceding stroke, partial pattern candidate storage means 706 for the separated result, It is provided with a partial pattern integrating unit 707 that integrates pattern candidates into one partial pattern.

When a signal indicating that handwritten data has been input to the handwritten pattern buffer 102 in the format shown in FIG. 14 is sent from the record / search control means 115 to the point data reading means 701, the point data reading means 701 reads the handwritten pattern buffer. Point data is read out from 102 in ascending order of point numbers in FIG.
Transfer to The stroke generating means 702 checks the pen up / down signal of the received point data. When the pen up / down signal of the point data up to immediately before is 0 and the pen up / down signal of the point data is 1, the point data is stored as a stroke start point. Thereafter, pen up / d of the point data continuously inputted
As long as the own signal is 1, the point data is recorded as a point in the stroke.

Furthermore, the pen up /
The down signal is 1, and the pen data of the point data is up / d.
When the own signal is 0, the immediately preceding point data is stored as the end point of the stroke, and data relating to the stroke is recorded in the form of the partial pattern data shown in FIG. First, the numbers in the order in which the strokes are detected are recorded as part numbers. The start point number and end point number obtained in the stroke detection stage are recorded in the start point number and end point number portions of the stroke number in the vertical column. Further, among the coordinates of all the points in the stroke, the X coordinate of the rightmost point is the upper right X coordinate, the Y coordinate of the uppermost point is the upper right Y coordinate, and the X coordinate of the leftmost point is the The lower left X coordinate and the Y coordinate of the lowest point are recorded as the lower left Y coordinate. When the detection of one stroke is completed, the strokes are sequentially detected by the above-described procedure while sequentially reading the point data, and the strokes are recorded in the form of a table shown in FIG. When all the points are read from the handwritten pattern buffer 102, the stroke generating means 702 stops.

When stroke data is output from the stroke generating means 702 in the format shown in FIG. 15, the inter-stroke time calculating means 703 and the inter-stroke moving amount calculating means 70
4 is activated. The inter-stroke time calculation means 703 includes:
The difference between the starting point number of the input stroke and the ending point number of the previously input stroke is obtained, and transferred to the separation determining unit 705. The inter-stroke movement amount calculating means 704 obtains the difference between the X coordinate of the start point of the input stroke and the X coordinate of the end point of the previously input stroke, and
Transfer to 05.

Further, a difference between the Y coordinate of the start point of the input stroke and the Y coordinate of the end point of the previously input stroke is obtained and transferred to the separation determining means 705. Separation determination means 7
05, the difference between the point numbers between the inputted strokes and X
Assuming that the difference between the coordinates and the difference between the Y coordinates are DT, DX, and DY, respectively, it is determined whether the stroke and the immediately preceding stroke are separated or integrated. In the present embodiment, the determination value D is calculated and obtained according to the following mathematical expression 1. When the determination value D is smaller than the threshold value T, the stroke is integrated with the immediately preceding stroke. If the previous stroke is already integrated with the previous stroke to form a partial pattern, the partial pattern and the stroke are integrated.

[0039]

(Equation 1) However, a, b, and c in Equation 1 are appropriately determined constants. In this embodiment, a, b, and c are 10, 10, respectively.
Although defined as 1, 1, this value is not essential and may be any value. When the separation determining unit 705 determines that the strokes are integrated, the partial pattern candidate storage unit 706 reads the stroke and the immediately preceding element, that is, the stroke, from the partial pattern storage table shown in FIG. Alternatively, a partial pattern is read.
Here, even if the immediately preceding element in the table of FIG. 15 is a stroke, it is also referred to as the immediately preceding partial pattern. The partial pattern candidate storage unit 706 reads the start point number of the immediately preceding partial pattern and the start point number of the stroke in the table of FIG. 15, and stores the smaller number as the start point number of the immediately preceding partial pattern. Further, the end point number of the immediately preceding partial pattern and the end point number of the stroke are read, and the larger number is stored as the end point number of the immediately preceding partial pattern. The upper right X coordinate of the immediately preceding partial pattern and the upper right X coordinate of the stroke are read, and the smaller value is stored as the upper right X coordinate of the immediately preceding partial pattern. The upper right Y coordinate of the immediately preceding partial pattern and the upper right Y coordinate of the stroke are read, and the smaller value is stored as the upper right Y coordinate of the immediately preceding partial pattern.

Similarly, the lower left X coordinate of the immediately preceding partial pattern and the lower left X coordinate of the stroke are read, and the larger value is stored as the lower left X coordinate of the immediately preceding partial pattern. The lower left Y coordinate of the immediately preceding partial pattern and the lower left Y coordinate of the stroke are read, and the larger value is stored as the lower left Y coordinate of the immediately preceding partial pattern. Finally, the element corresponding to the stroke is deleted from the table of FIG. Stroke generating means 7
02 reads out all the point data stored in the handwritten pattern buffer 102, generates a stroke, and transfers it to the separation determination unit 705. The separation determining unit 705 determines whether a partial pattern is separated between consecutive strokes, and stores a set of strokes determined not to be separated by the partial pattern candidate storage unit 706.
After that, the partial pattern integration unit 707 is activated.

In the partial pattern integration means 707, FIG.
Is determined for each individual partial pattern that is an element of the above table, and the integration process is performed. The two partial patterns whose numbers are adjacent are taken out in ascending order of the partial pattern number,
Perform integration judgment. In the integration determination, first, the partial pattern and the immediately preceding partial pattern are extracted. The immediately preceding partial pattern refers to a partial pattern having the largest partial number among partial patterns whose partial number is smaller than the partial number of the partial pattern. The integration determination will be described for the case of horizontal writing. The condition for the integration determination is at least from the upper right X coordinate of the partial pattern to the lower left X of the immediately preceding partial pattern.
The value obtained by subtracting the coordinates is larger than the threshold value.

Here, the threshold value is one third of the difference between the lower left X coordinate and the upper right X coordinate of the immediately preceding partial pattern, but may be any value. If this condition is satisfied, the partial pattern and the immediately preceding partial pattern are integrated. First, the starting point number of the immediately preceding partial pattern and the starting point number of the partial pattern in the table of FIG. 4 are read, and the smaller number is stored as the starting point number of the immediately preceding partial pattern. Also,
The ending point number of the immediately preceding partial pattern and the ending point number of the partial pattern are read out, and the larger number is stored as the ending point number of the immediately preceding partial pattern. The upper right X coordinate of the immediately preceding partial pattern and the upper right X coordinate of the partial pattern are read, and the smaller value is stored as the upper right X coordinate of the immediately preceding partial pattern. The upper right Y coordinate of the immediately preceding partial pattern and the upper right Y coordinate of the partial pattern are read, and the smaller value is stored as the upper right Y coordinate of the immediately preceding partial pattern. Similarly, the lower left X coordinate of the immediately preceding partial pattern and the lower left X coordinate of the partial pattern are read, and the larger value is stored as the lower left X coordinate of the immediately preceding partial pattern. The lower left Y coordinate of the immediately preceding partial pattern and the lower left Y coordinate of the partial pattern are read, and the larger value is stored as the lower left Y coordinate of the immediately preceding partial pattern. Finally, the element corresponding to the partial pattern is deleted from the table shown in FIG.

The above-described integration determination and integration processing are performed in the order of the part numbers, and the integration determination and the integration processing are performed up to the partial pattern having the largest part number. After the completion of the integration determination and the integration processing of all the partial patterns, the resulting partial pattern is transferred to the feature extracting unit 106 in FIG. more than,
Although the partial pattern separating means 103 has been described in detail in one embodiment, the means for generating a partial pattern in the present invention is not limited to the above. For example, a partial pattern separating unit using the frame used in the above description of the related art may be used. In the present embodiment, the stroke numbers are numbered in the order in which the strokes are input. However, once the stroke data is generated as shown in FIG. 15, the stroke numbers are calculated based on the upper right X coordinate and the upper right Y coordinate of the stroke. The above processing may be performed after rearranging and numbering.
Similarly, based on the lower left X coordinate and lower left Y coordinate of the stroke,
The above processing may be performed after rearranging and numbering.

The operation of the feature extracting means 106 will be described with reference to FIG. The data of the partial pattern output from the partial pattern separating means 103 is input to the feature extracting means 106 through the terminal 801 for each partial pattern, and the point thinning means 802, the start point end point direction measuring means 81
0, sent to the vertical and horizontal size measuring means 811.

First, the point thinning means 802 reads the start point number and the end point number from the data of one partial pattern in the partial pattern set recorded in the format shown in FIG. All the point data between the start point number and the end point number are read out from those recorded in the handwritten pattern buffer 102 in the format of FIG. First, the distance between the coordinates of the starting point and the coordinates of the next point is calculated, and if the distance is smaller than the thinning distance threshold value J, the next point is deleted. Further, the next point is read, and the distance from the immediately preceding point, in this case, the starting point is calculated. If the distance is smaller than the thinning distance threshold value J, the next point is deleted. This is repeated while reading out points.

If the point-to-point distance is larger than the threshold value J, the most recently read point is set immediately before the next point and the next point is read out.
The point removal process is repeated. The points that have not been removed are determined by detecting the direction between adjacent points 803, the number of strokes measuring means 805,
The data is sequentially transferred to the total stroke length measuring means 806. The direction detecting means between adjacent points 803 calculates and obtains the direction between the two points by using the coordinates of the two points that are successively read. The coordinates of the smaller number of the two points are (x
1, y1) and the coordinates of the other point are (x2, y2)
And At this time, the direction between the two points is calculated and calculated using the mathematical formula shown in Equation 2 below.

[0047]

(Equation 2) Here, A is a real number from 0 to 2π. The direction A obtained by the calculation is transferred to the direction counting means 804. Also,
The direction of a fixed number of points from the start point is the start point direction measuring means 808.
Transfer to In the present embodiment, the fixed number is five, but this value is not essential and may be any number. The directions of a fixed number of points immediately before the end point are transferred to the end point direction measuring means 809. In the present embodiment, the fixed number is five, but this value is not essential and may be any number. The direction counting unit 804 classifies the received direction A into a class of the number of directions N. An N-class histogram initially set to 0 is prepared for each partial pattern, and each time the direction A is received, the direction A is set to π * i /
Search for i that is greater than N and less than π * (i + 1) / N, i
Add 1 to the th histogram. At the same time, 1 is added to the point counter. In the present embodiment, the number of classes N used here is eight, but any number may be used. At the stage where all points in one partial pattern are transferred from the adjacent point direction detecting means 803 and the direction counting is completed,
A value is obtained by dividing each histogram value of the N class by the total number of points, and is output from the terminal 812.

The starting point direction measuring means 808 calculates and calculates an average value of the five directions received from the direction detecting means 803 between adjacent points, and outputs the average value from the terminal 812. The end point direction measuring means 809 calculates and calculates the average value of the five directions received from the direction detecting means 803 between the adjacent points.
Output from 2.

In the stroke number measuring means 805, the data of the points sequentially received from the point thinning means
By examining the down signal, the number of pen downs, that is, the number of interruptions to 0 after a continuation of 1 is counted. The result is output from the terminal 812 as the number of strokes and transferred to the average stroke length measuring means 807. The total stroke length measuring means 806 calculates the distance between the two points from the coordinates of two consecutive points where the pen up / down signal is 1 based on the data of the points transferred from the point thinning means 802. Then, the accumulated value is obtained, output from the terminal 812, and transferred to the average stroke length measuring means 807. In addition to this, the total stroke length measuring means 806
Then, the total stroke length may be obtained by counting the number of points where the pen up / down signal is 1 in the data of the points transferred from the point thinning means 802. In the average stroke length measuring means 807, the total stroke length measuring means 8
A value obtained by dividing the value obtained from the number 06 by the value obtained from the stroke number measuring means 805 is output from the terminal 812 as an average stroke length.

The start / end point direction measuring means 810 includes a terminal 801.
The coordinates of the two points are read from the handwritten pattern buffer 102 from the start point number and the end point number of the partial pattern input from, and the coordinates of the two points are (x1, y1), (x
(2, y2) is calculated using the formula of Equation 2 as the direction between the start point and the end point as A, and is output from the terminal 812. The vertical / horizontal size measuring means 811 calculates the difference between the lower left X coordinate and the upper right X coordinate of the partial pattern input from the terminal 801 as the horizontal width, and calculates the difference between the lower left Y coordinate and the upper right Y coordinate of the partial pattern as the vertical width. Output from

The characteristic measurement values of each partial pattern output from the terminal 812 are stored in the format shown in the table of FIG. The first switch 107 is connected to the feature / pointer storage means 109 at the time of storing the handwritten pattern in accordance with the signal from the record / search control means 115.
The data of the partial pattern in the format shown in FIG. When searching for a handwritten pattern, the first switch 10
7 is connected to the feature and pointer buffer 111, the partial pattern data of the format shown in FIG.
1 is recorded.

At the time of retrieval, the feature and pointer buffer 1
The feature matching unit 112 is activated at the stage when the feature of the partial pattern is recorded in 11. The operation of the feature matching unit 112 will be described with reference to FIGS. FIG. 5 shows the feature matching means 11.
2 is a detailed block diagram of FIG.
2 is a flowchart showing the operation of the characteristic error calculation means 904 in FIG. The first feature buffer 903 reads and holds the feature from the feature and pointer buffer 111 by designating the part number of the partial pattern. Similarly, the second feature buffer 902 designates the part number of the partial pattern and sends the feature and pointer storage means 10 through the feature reading means 110.
9 and read and hold the feature.

The feature combining means 901 combines the features of a plurality of partial patterns to generate one feature. In order to combine the features generated by the direction counting unit 804, the sum of the direction counts of a plurality of features is obtained for each direction, and the sum is used as the feature value of each direction of the combined feature. When combining the features in the direction of the starting point generated by the starting point direction measuring means 808, the starting point of the partial pattern having the smallest point number among the starting points of the plurality of partial patterns to be synthesized is determined as the starting point of the partial pattern to be synthesized. And the starting point direction of the starting point is the starting point direction of the combined partial pattern.

When combining the features in the direction of the end point selected and generated by the end point direction measuring means 809, the end point of the partial pattern having the largest point number among the end points of the plurality of partial patterns to be synthesized is synthesized. As the end point of the partial pattern to be combined, the end point direction of that point is set as the end point direction of the combined partial pattern.

For the synthesis of the start point end point direction generated by the start point end point direction measuring means 810, the coordinates of the point are obtained from the point number of the start point of the previously obtained partial pattern to be synthesized. From the point number of the end point of the partial pattern to be obtained previously, the coordinates of that point are obtained, and the direction between the points is calculated and calculated according to the equation shown in Equation 2, and the combined partial pattern is obtained. From the start point to the end point. In addition, in order to combine the number of strokes measured by the number-of-strokes counting means 805, a total sum of the number of strokes of a plurality of partial patterns to be combined is obtained and set as the number of strokes.

To combine the total stroke lengths generated by the total stroke length measuring means 806, the sum of the total stroke lengths of a plurality of partial patterns to be combined is obtained.
Total stroke length. In addition, in order to combine the average stroke length generated by the average stroke length measuring means 807, the combined total stroke length obtained above is divided by the combined stroke length, and the quotient is calculated as the combined average. Stroke length. In addition, in order to combine the features generated by the vertical and horizontal size measuring means 811, the smallest upper right X coordinate is determined to be X1 and the smallest upper right Y coordinate is determined to be Y1 among a plurality of combined partial patterns. The largest lower-left X coordinate is determined as X2, the largest Y coordinate is determined as Y2, the horizontal size is calculated by X2-X1, the vertical size is calculated by Y2-Y1, and each is calculated. The horizontal size and vertical size of the partial pattern to be combined are set.

The feature error calculating means 904 first calculates the error (difference between two features) VD between the partial patterns by the following equation (3).
The total is calculated with the load according to the formula shown in (1).

[0058]

(Equation 3) Here, fi is the i-th kind of feature extracted from the feature and the search key pattern read from the pointer buffer 111. In the present embodiment, eight direction counts, start direction, end direction, stroke number , The stroke length, the average stroke length, and the start point end point direction are shown in order. That is, i is a natural number from 1 to 14. Similarly, gi is the i-th type feature extracted from the feature and the accumulation pattern read from the pointer storage means 109. vi is a count for the load, which is 1 in this embodiment, but may be any value. In addition, Equation 3
| X | represents the absolute value of x. Further, as another embodiment, the error VD between the partial patterns can be calculated according to the mathematical expression shown in the following expression 4 instead of the above expression 3.

[0059]

(Equation 4) Here, Hf and Vf represent the horizontal size and vertical size of the partial pattern of the search key pattern. As still another embodiment, the error VD between the partial patterns can be calculated using the mathematical formula shown in the following Expression 5 instead of the above Expression 3.

[0060]

(Equation 5) Here, Hg and Vg represent the horizontal size and vertical size of the partial pattern of the storage pattern. As still another embodiment, instead of Hf * Vf in Equation 4, Hf,
Sqrt (Hf * Vf), Sqrt (Hg * Vg), or Sqrt (Hf
* Vf * Hg * Vg) to calculate the error VD between partial patterns.
Can also be calculated. Here, Sqrt (x) represents the square root of x.

A series of partial patterns corresponding to all the partial patterns belonging to the retrieval key pattern are read out from the accumulated patterns by using the calculation means of the error VD between the partial patterns, and the sum of the errors between the individual partial patterns is obtained. It is calculated as an error between the search key pattern and the searched pattern. In another embodiment, the error SD of the entire pattern is calculated according to the following equation (6), with the error of each partial pattern between the key pattern and the corresponding searched pattern as VD (j). You can ask.

[0062]

(Equation 6) Here, m is the total number of partial patterns used correspondingly in the search key pattern. VD (j), Hf
(j) and Vf (j) are the error, horizontal size, and vertical size of the j-th partial pattern in the search key pattern.
Further, in another embodiment, Hf (j) * Vf (j)
Alternatively, Sqrt (Hf (j)) * Sqrt (Vf (j)) or Hg (j) * Vg (j) or Hf (j) can be used.

In the process of calculating the error of the partial pattern in the search key pattern, the characteristic error calculating means 904 obtains a partial pattern corresponding to the accumulated pattern corresponding to the partial pattern and stores it in the corresponding partial storage means 905. .

FIG. 7 is a diagram showing a case where the partial patterns F1 to F8 of the search key pattern are matched with the partial patterns T1 to T8 of the storage pattern. Round 1001
Indicates that the partial pattern F3 and the partial pattern T3 are associated with each other. Similarly, circle 1002 is
This indicates that the partial pattern F4 and the partial pattern T4 are associated with each other. The oval 1005 is a composite pattern of the partial pattern F3 and the partial pattern F4 and a composite pattern of the partial pattern T3 and the partial pattern T4.
Indicates that they are associated with each other. That is, there are two methods for associating the partial patterns F3 and F4 with the partial patterns T3 and T4, and it is necessary to select an appropriate one of them. The characteristic error calculation means 904
Run in

The operation of the feature error calculating means 904 and the operation of the elements in the feature matching means 112 will be described with reference to FIG.
First, in the feature error calculation means 904, a buffer D '(i, j) is prepared in step S2 immediately after step S1 at the beginning of the process, and all elements are cleared to zero.
Here, i is an index indicating a partial pattern in the search key pattern, and takes a natural number from 1 to the total number of partial patterns. Also, j is an index indicating a partial pattern in all the accumulated patterns, and takes a natural number from 1 to the number of all partial patterns. J is initially set to 1. D 'is called deep prime. Next, step S3
In step S4, i, i 'and j' are initialized. In step S5, it is determined whether or not D '(ii', j-j ') is smaller than the threshold value. If smaller than the threshold value, the process proceeds to step S6. If larger, the process proceeds to step S9. , D (i, i ', j, j') to D '(ii', j
−j ′).

In step S6, from the (j-j ')-th partial pattern to the j-th partial pattern in the accumulated pattern are synthesized in accordance with the above-described method of obtaining the characteristic when the partial patterns are synthesized. The characteristic is obtained as P.
However, when (j−j ′) becomes smaller than 1,
A feature is obtained by combining the first partial pattern to the j-th partial pattern to obtain a feature. Next, at step S7, similarly, a feature is obtained by combining the (i-i ')-th partial pattern to the i-th partial pattern in the search key pattern to obtain a feature. However, (ii ') is 1
If it becomes smaller, the features from the first partial pattern to the i-th partial pattern are synthesized to obtain a feature.
And

Further, in step S8, according to the method for calculating the error between the corresponding partial patterns described above, the error between the feature P and the feature Q is obtained and D 'already calculated.
(I−i ′, j−j ′) and add D (i−i ′, j−
j ′). i ′ is 1 in order from 0 to maxi ′
While increasing, the processing from step S5 to step S8 is repeated to obtain D (i, i ', j, j'). In the present embodiment, maxi '(max-eye prime) is set to 3, but may be any value. Further, j ′ is set to 0
To maxj ′, the processing from step S5 to step S10 is repeated while incrementing by 1 to D (i,
i ', j, j').

Then, the process proceeds to a step S12, wherein the previously obtained D
Of (i, i ', j, j'), the one with the smallest value after changing i 'and j' is selected and substituted for D (i, j). At that time, D (i, i ', j,
j ') from (i' and j '), (ii'-1) and (j-j'
-1) to obtain the pointer storage units PI (i, j), PJ
(I, j). Next, the process proceeds to step S13, where i
Step S while increasing by 1 until becomes 1 from maxi
4 to step S12 are repeated to obtain D (i,
j) is obtained. In step S14, D (maxi, j)
Is smaller than the threshold value T, it is assumed that a pattern matching the search key pattern has been detected from the accumulated pattern. In that case, the process proceeds to step 15 and PI
(Maxi, j) and PJ (maxi, j) are read out and assigned to pointer variables pi and pj and stored in the corresponding part storage means 905. Further, PI (pi, p
j) and PJ (pi, pj) are read and assigned to pi and pj and stored in the corresponding part storage means 905. This is repeated until pi becomes smaller than 1. If D (maxi, j) is larger than the threshold value T in step 14, the process proceeds to step S16, and the processing from step S3 to step S16 is repeated while j is incremented by 1 from 1 to maxj. Here, maxj is the number of all partial patterns included in the accumulation pattern. After ending this, the process proceeds to step S17 and ends.

As a result, the corresponding portion storage means 905 stores
A plurality of sets of partial patterns corresponding to the search key pattern are recorded and transferred to the handwritten pattern reading means 113 through the terminal 906.

The handwritten pattern reading means 113 sets the set of partial pattern numbers (0, pj) and (maxi, pj ') in one of the received plural stored patterns.
From the pj and pj ′ of
The start point number of the pj-th partial pattern and the end point number of the pj'-th partial pattern stored therein are read as qj and qj '. The data from the qj-th point to the qj-th point is read out and displayed on the handwritten pattern display means 114. In another embodiment of the handwritten pattern storage means 108, the handwritten pattern storage means 1
08, the page number of the qj-th point is read, and all the point data in the page are read and displayed on the handwritten pattern display means 114. Furthermore, the data from the qj-th point to the qj-th point stored in the handwritten pattern storage means 108 is read out, and the color or size is displayed on the handwritten pattern display means 114 with the thickness changed.

Next, a second embodiment of the present invention will be described with reference to FIG. The difference between the present embodiment and the above-described first embodiment is that a handwritten pattern matching unit 111 is provided after the feature matching unit 112 in the first embodiment.
2 has. After the operation of the feature matching unit 112, the corresponding part storage unit 905 sets the part numbers of the partial patterns in the accumulation pattern corresponding to the partial patterns in the search key pattern and stores them as a set of (pi, pj). I have. Assume that a total of M sets are stored, and for the sake of convenience, the set of partial patterns is herein referred to as (pi (m), pj
(M)). Here, m is a natural number from 1 to M.

Therefore, the handwritten pattern matching means 1112 firstly stores the point data from the start point of the (pj (1) +1) th partial pattern in the storage pattern to the end point of the pj (2) th partial pattern in the handwritten pattern. Reading is performed through the reading unit 113. Further, point data from the start point of the (pi (1) +1) th partial pattern in the search key pattern to the end point of the pi (2) th partial pattern is read from the handwritten pattern buffer 102. The two point sequences are matched, and the matching degree is obtained and added to the cumulative matching degree R. Next, the (pj (2)
Point data from the start point of the (+1) th partial pattern to the end point of the pj (3) th partial pattern is read through the handwritten pattern reading means 113.

Further, the (pi) in the search key pattern
From the start point of the (2) +1) th partial pattern,
(3) The point data up to the end point of the third partial pattern is read from the handwritten pattern buffer 102. The two point sequences are matched, and the matching degree is obtained and added to the cumulative matching degree R. By repeating the above processing, all pi (m) and pj
Assuming that R obtained after performing (m) is the final cumulative matching degree, only when the cumulative matching degree exceeds a threshold value, the pattern of the point sequence read from the handwritten pattern storage unit 108 is It is displayed on the handwritten pattern display means 114. If the cumulative consistency falls below the threshold,
Another matching partial pattern set transferred from the feature matching unit 112 is read, and the same processing is repeated. Here, any processing may be used to determine the degree of consistency between the two point sequences. However, by using one point sequence as an input pattern and the other point sequence as a recognition dictionary pattern, "Hashimoto Shin-ichiro, Introduction to Recognition, published by Ohmsha, March 1982, p.
p. 126 to 135 ".

In another embodiment of the handwritten pattern storage means 108, if the cumulative consistency exceeds the threshold value, the (pj) stored in the handwritten pattern storage means 108
The page number of the (1) +1) th point is read, all the point data in the page are read, and displayed on the handwritten pattern display means 114. Further, the (pj (1) +) stored in the handwritten pattern storage means 108
1) The data from the first point to the pj (M) th point is read out and displayed on the handwritten pattern display means 114 by changing the color, size or thickness.

Next, a third embodiment of the present invention will be described with reference to FIG. The difference between the present embodiment and the above-described first embodiment is that a second recognition unit 1203 is provided after the feature matching unit 112 in the first embodiment, and the handwriting pattern reading unit 113 After that, it has a first recognizing means 1201 and a character comparing means 1203 having inputs from the two. After the operation of the feature matching unit 112, the corresponding part storage unit 905 stores the part numbers of the partial patterns in the accumulation pattern corresponding to the partial patterns in the search key pattern as a set of sets (pi, pj). . It is assumed that a total of M sets are stored, and for convenience, a set of partial patterns is referred to as (p
i (m)) and pj (m)). Here, m is a natural number from 1 to M.

First, the first recognizing means 1201 converts point data from the start point of the (pj (1) +1) th partial pattern in the accumulation pattern to the end point of the pj (2) th partial pattern into a handwritten pattern. Reading is performed through the reading unit 113. The read point sequence data is recognized using an online character recognition method, and as a result, a set of a plurality of character codes and their certainty factors is obtained. The method of recognizing the read dot sequence may be any method. For example, see Shinichiro Hashimoto, Introduction to Character Recognition, published by Ohmsha, March 1982,
pp. 126 to 135 ".

Next, the (pj (2) +
Point data from the start point of the 1) th partial pattern to the end point of the pj (3) th partial pattern is read through the handwritten pattern reading means 113. The read point sequence data is recognized using an online character recognition method, and as a result, a set of a plurality of character codes and their certainty factors is obtained. The above process is repeated for all pj (m) to obtain a set of (C1 (i, j), W1 (i, j) and transfer it to the character comparison means 1203. The i is the i-th. Indicates that the recognition result is obtained by performing the recognition processing on the partial pattern, j indicates the j-th candidate of the recognition processing, and C1 (i, j) indicates the character of the recognition result. , W1 (i, j) indicate the certainty factor of C1 (i, j).
Similarly, the second recognizing means 1202 calculates point data from the start point of the (pi (1) +1) th partial pattern in the search key pattern to the end point of the pi (2) th partial pattern. Read from the handwritten pattern buffer 102. The read point sequence data is recognized using an online character recognition method, and as a result, a set of a plurality of character codes and their certainty factors is obtained.

Next, the (pi) in the search key pattern
From the start point of the (2) +1) th partial pattern,
(3) The point data up to the end point of the third partial pattern is read from the handwritten pattern buffer 102. Recognize the read point sequence data using online character recognition method,
As a result, a set of a plurality of character codes and their certainty factors is obtained. The above processing is repeated for all pi (m) to obtain a set of (C2 (i, j), W2 (i, j)) and transfer it to the character comparison means 1203. i indicates that the result is obtained by performing the recognition process on the i-th partial pattern. Also, j indicates that it is the j-th candidate for the recognition process. C2 (i, j) indicates the character of the recognition result, and W2 (i, j) indicates the certainty factor of C2 (i, j).

In the character comparing means 1203, first, (C1
(I, j), W1 (i, j)) and (C2 (1,
j ′), W2 (1, j ′)) and compare them.
Here, j and j 'are arbitrary natural numbers from 1 to the maximum number of candidates. Although any means may be used for the comparing means, for example, a method as disclosed in JP-A-8-305722 can be used. That is, C1 (i, j) = C2 (1,
j ′) and W1 (i, j) * W2 (1, j ′)
Is detected, and W1 (i, j) * W2 (1,
j ′) is substituted for R. Similarly, (C1 (2, j), W
1 (2, j)) and (C1 (2, j ′), W1 (2,2)
j ′)) is taken out and compared, and C1 (2, j) = C2
(2, j ') and W1 (2, j) * W2 (2,
j ′) is maximized, and W1 (2, j) * W2
(2, j ′) is added to R. This is all (C1
(I, j), W1 (i, j)) and (C2 (i, j,
j ′), W2 (i, j ′)), and compare the final R with a threshold. If R is larger than the threshold value, it is determined that the accumulated pattern corresponding to the search key pattern has been detected, and the point sequence read by the handwritten pattern reading means 113 is displayed on the handwritten pattern display means 114. . If the value R falls below the threshold value, another matching partial pattern set transferred from the feature matching unit 112 is read out and the same processing is repeated.

In another embodiment of the handwritten pattern storage means 108, if the cumulative consistency exceeds a threshold value, the (pj) stored in the handwritten pattern storage means 108
The page number of the (1) +1) th point is read, all the point data in the page are read, and displayed on the handwritten pattern display means 114. Further, the (pj (1) +) stored in the handwritten pattern storage means 108
1) The data from the first point to the pj (M) th point is read out and displayed on the handwritten pattern display means 114 by changing the color, size or thickness.

[0081]

According to the present invention, the matching is checked after being separated into partial patterns in advance, which is much faster than checking the matching comprehensively for each stroke. Can be searched and displayed. For example, assuming that one partial pattern is composed of five strokes, the speed can be increased by a factor of 25 by integrating the strokes into the partial patterns in advance.

Further, according to the present invention, when a stroke is integrated into a partial pattern, one character is separated into two or integrated with another character, and a portion that is not balanced with the pattern at the time of accumulation is obtained. Even when a pattern is generated, multiple partial patterns are integrated to obtain a corresponding partial pattern before matching, so that the pattern entered at the time of accumulation and the pattern entered at the time of retrieval are Even if there is a discrepancy due to, for example, radical spacing, it can be correctly detected.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of the present invention.

FIG. 2 is a block diagram showing a configuration of a partial pattern separating unit.

FIG. 3 is a block diagram illustrating a configuration of a feature extracting unit.

FIG. 4 is a diagram illustrating a data format of a partial pattern after features are extracted.

FIG. 5 is a block diagram illustrating a configuration of a feature matching unit.

FIG. 6 is a flowchart showing a procedure of processing by a feature matching unit.

FIG. 7 is a diagram illustrating matching between a partial pattern and a combined partial pattern.

FIG. 8 is a block diagram showing a configuration of another embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of another embodiment of the present invention.

FIG. 10 is a diagram illustrating an example of a handwritten pattern.

FIG. 11 is a diagram illustrating an example of a result obtained by separating a handwritten pattern into partial patterns.

FIG. 12 is a diagram illustrating an example of a result obtained by separating a handwritten pattern into partial patterns.

FIG. 13 is a block diagram showing a configuration of a conventional handwritten character storage and retrieval device.

FIG. 14 is a diagram illustrating a format of point data.

FIG. 15 is a diagram illustrating the format of partial pattern and stroke data.

FIG. 16 is a diagram illustrating the format of partial pattern data in a conventional handwritten pattern storage and retrieval device.

[Explanation of symbols]

 Reference Signs List 101 Handwritten pattern input means 102 Handwritten pattern buffer 103 Partial pattern separation means 105 Second switch 106 Feature extraction means 107 First switch 108 Handwritten pattern storage means 109 Features and pointer storage means 110 Features readout means 111 Features and pointer buffer 112 Features Matching means 113 Handwritten pattern reading means 114 Handwritten pattern display means 115 Record / search control means 701 Point data reading means 702 Stroke generating means 703 Interstroke time calculating means 704 Stroke moving amount calculating means 705 Separation determining means 706 Partial pattern candidate storage means 707 Partial pattern integration means 708 Terminal 801 Terminal 802 Point thinning means 803 Direction detection means between adjacent points 804 Direction counting means 805 Number measuring means 806 total stroke length measuring means 807 average stroke length measuring means 808 start point direction measuring means 809 end point direction measuring means 810 start point end point direction measuring means 811 vertical and horizontal size measuring means 812 terminal 901 feature synthesizing means 902 second characteristic buffer 903 First feature buffer 904 Feature error calculation means 905 Corresponding partial storage means 906 Terminal 1112 Handwritten pattern matching means 1201 First recognition means 1202 Second recognition means 1203 Character comparison means

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] December 7, 1998

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claim 2

[Correction method] Change

[Correction contents]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction contents]

Further, according to the present invention, the feature matching unit combines a plurality of partial patterns to obtain a feature of the combined pattern, and uses a feature of the original partial pattern or a feature of the combined pattern. Thus, there is provided a handwritten pattern storage and retrieval apparatus characterized by comprising a feature error calculating means for performing feature matching.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0032

[Correction method] Change

[Correction contents]

Next, the operation will be described. Handwritten pattern is inputted to the input unit 101, the handwritten pattern output therefrom is output, for example, format as shown in FIG. 14. Figure 1
4 will be described. Pen moving coordinate is sampled every predetermined time intervals, and outputs the data as a vertical column of Figure 14. One sampled data includes a serial number of the point, a page number on which the data is described, an X coordinate and a Y coordinate in the page, and a pen up indicating whether the pen tip is in contact with the tablet.
/ Down signal. When the pen tip is in contact with the tablet, 1 is output as the pen up / down signal, that is, a pen down signal is output. When the pen tip is not in contact with the tablet, 0 is output as the pen up / down signal , that is, the pen is down. An up signal is output.

Claims (12)

[Claims] 1. A handwritten pattern storage and retrieval device comprising a handwritten pattern input means and a handwritten pattern storage means for retrieving a stored pattern similar to a search key pattern input from the handwritten pattern input means. Means, partial pattern feature extracting means, feature matching means for matching the features of the stored partial patterns and the features of the search key pattern, and means for reading out a handwritten pattern corresponding to the matched partial pattern. An apparatus for storing and retrieving handwritten patterns, wherein the apparatus is configured. 2. A feature matching unit that combines a plurality of partial patterns to obtain a feature of a combined pattern, and performs feature matching by using a feature of an original partial pattern or a feature of a combined pattern. 2. The handwritten pattern storage and retrieval apparatus according to claim 1, wherein the apparatus is provided with matching executing means. 3. The feature synthesizing means includes means for calculating a sum of partial pattern data from features of a plurality of partial patterns and generating a feature of the combined pattern as a combined feature value. Handwritten pattern storage and retrieval device. 4. The method according to claim 1, wherein the plurality of partial pattern data includes direction count data, start point direction data, end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data. 4. The apparatus for storing and retrieving handwritten patterns according to claim 3, wherein: 5. A handwritten pattern storage and retrieval device comprising a handwritten pattern input means and a handwritten pattern storage means for retrieving a stored pattern similar to a search key pattern input from said handwritten pattern input means. Means, partial pattern feature extracting means, feature matching means for matching the features of the stored partial patterns and the features of the retrieval key pattern, means for reading a handwritten pattern corresponding to the matched partial pattern, An apparatus for storing and retrieving a handwritten pattern, comprising: a handwritten pattern matching means for matching a set handwritten pattern with a search key pattern. 6. A feature matching unit that combines a plurality of partial patterns to obtain a feature of a combined pattern, and performs feature matching using the features of the original partial pattern or the features of the combined pattern. 6. The handwritten pattern storage and retrieval apparatus according to claim 5, wherein the apparatus is provided with matching executing means. 7. The feature synthesizing means includes means for calculating a sum of partial pattern data from characteristics of a plurality of partial patterns and generating a characteristic of the synthetic pattern as a synthetic characteristic value. Handwritten pattern storage and retrieval device. 8. The method according to claim 1, wherein the plurality of partial pattern data are composed of direction count data, start point direction data, end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data. The handwritten pattern storage and retrieval device according to claim 7, characterized in that: 9. A handwritten pattern storage and retrieval device which has a handwritten pattern input means and a handwritten pattern storage means and searches for a stored pattern similar to a search key pattern input from the handwritten pattern input means. Means, partial pattern feature extracting means, feature matching means for matching the features of the stored partial patterns and the features of the search key pattern, means for reading out a handwritten pattern corresponding to the matched partial pattern, Recognition means for executing recognition using the result of separating the output handwritten pattern into partial patterns by the partial pattern separation means as input, and inputting the result of separating the search key pattern into partial patterns by the partial pattern separation means. Recognizing means for performing recognition, and the read handwritten pattern An apparatus for storing and retrieving handwritten patterns, comprising: a character comparing unit configured to compare a result of recognition with a result of recognizing the search key pattern. 10. The feature matching unit combines a plurality of partial patterns to obtain a feature of the combined pattern, and performs feature matching using the features of the original partial pattern or the features of the combined pattern. 10. The handwritten pattern storage and retrieval apparatus according to claim 9, wherein the apparatus is provided with matching executing means. 11. The feature synthesizing means includes means for calculating a sum of partial pattern data from characteristics of a plurality of partial patterns and generating a characteristic of the synthetic pattern as a synthetic characteristic value. Handwritten pattern storage and retrieval device. 12. The method according to claim 1, wherein the plurality of partial pattern data includes direction count data, start point direction data, end point direction data, stroke number data, stroke length data, average stroke length data, and start point end point direction data. The handwritten pattern storage and retrieval device according to claim 11, characterized in that: