JP2007213433A - Character retrieving apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character retrieving apparatus which retrieves a character by a partial pattern even when a plurality of images are inputted by one stroke. <P>SOLUTION: The character retrieving apparatus is provided with: a stroke similarity degree calculation part 13 for referring to a stroke dictionary storage part 17 in which at least dictionary strokes and stroke order data are registered and finding out the similarity degree of an input stroke to a dictionary stroke string; a similarity degree table preparation part 14 for preparing a similarity degree table for the combinations of input strokes included in one input pattern with the dictionary stroke string of one character or symbol; a similarity degree sum calculation part 15 for finding out the total sum of maximum similarity degrees for respective input strokes from the similarity degree table; a penalty value subtraction part 16 for subtracting a prescribed penalty value from the sum of maximum similarity degrees to find out a pattern matching degree when a prescribed penalty condition is formed by referring to the similarity degree table; and a retrieved result presentation part 18 for presenting characters or symbols as retrieved results in the descending order of pattern matching degrees. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a character search device that searches for a character or symbol based on a pattern input by handwriting using a pointing device capable of handwriting input, and in particular, searches for a character or symbol that matches at least part of the input pattern. The present invention relates to a character search apparatus that can efficiently perform the above.

  2. Description of the Related Art Conventionally, information processing apparatuses that can input characters and symbols by hand using a pointing device capable of handwriting input such as a pen tablet or a mouse are widely known. In such an information processing apparatus, by comparing the input pattern with the character or symbol pattern registered in the dictionary, the handwritten input pattern is the character or symbol registered in the character dictionary. Various methods have conventionally been proposed for handwritten character recognition (also referred to as handwritten character search) for determining which one matches or is similar.

  Also, when the user wants to input kanji characters, kanji characters have more strokes than kana characters, so even if input of the entire character is not completed, characters are searched based on partial patterns and character candidates with high similarity Can be presented to the user, which is advantageous in that it saves the user's input and allows the user to input kanji even if the user does not remember the entire configuration of the kanji. FIG. 7 shows an example of character search using partial patterns. As shown in FIG. 7, when the user inputs at least a part of the kanji, character candidates including a pattern similar to the input partial pattern are displayed, so that the user can input from the displayed character candidates. Just select the kanji you tried.

  For example, the following techniques are known as conventional handwritten character recognition or handwritten character search methods.

  For example, Patent Document 1 (Japanese Patent Application Laid-Open No. 7-57056) discloses an online character recognition device that enables high-speed and high-accuracy character recognition by recognizing handwritten characters in order of strokes, partial patterns, and characters. ).

  Also, a similarity matrix is generated by calculating the similarity between the input and each stroke of the dictionary, the element having the maximum similarity in each column is uniquely determined for each row, and the sum of the maximum similarity is calculated. A character recognition method for obtaining the similarity between characters by taking them and recognizing the characters is disclosed in Patent Document 2 (Japanese Patent No. 3155777). According to this character recognition method, it is possible to perform a character search independent of the stroke order.

  Furthermore, the technology that achieves character recognition that allows stroke order differences and strokes to be written continuously by taking rough correspondence between strokes and strokes, and then aligning strokes so that there is no duplication. It is disclosed in Patent Document 1.

  Further, Japanese Patent Laid-Open No. 8-123903 discloses a technique for performing partial pattern character search by performing pattern recognition on a partial pattern and searching for a kanji from a partial pattern and kanji correspondence table. Has been.

Further, a method of performing partial pattern character search by sequentially matching the strokes of the input pattern and the dictionary pattern one by one using a dictionary of stroke shape and positional relationship between strokes is disclosed in JP-A-9-190499. ). Since this method is not limited by the types of partial patterns that can be searched, a flexible search is possible.
JP 7-57056 A Japanese Patent No. 3155777 JP-A-8-123903 JP-A-9-190499 Kazutaka Odaka, Toru Wakahara, Michio Umeda, “Online Character Recognition Independent of Stroke Number and Stroke Order by Selective Stroke Combination”, IEICE Transactions, May 1983, J66-D, No.5, p.593- 600

  In the case of handwritten character input, especially in the case of a complex kanji with a large number of strokes, it may be difficult to write the entire character correctly. In many cases, a partial pattern originally composed of a plurality of images is input in one stroke by successive writing. Therefore, it is preferable that the character search by the partial pattern is possible even when the entire character is not written correctly or when a plurality of images are input in one stroke.

  For example, in the search method described in Patent Document 3, a character that is not in the correspondence table between partial patterns and characters cannot be searched, and even if a pattern that is not in the partial pattern dictionary is written, it cannot be searched. Also, it is difficult to create a complete correspondence table or partial pattern dictionary.

  In the method described in Patent Document 4, since the stroke of the input pattern is collated with the stroke of the dictionary pattern one by one, when a plurality of images are input with one stroke, a correct search cannot be performed.

  As a character recognition method that copes with the continuous writing of a plurality of images, the method of Non-Patent Document 1 has been proposed. This method is based on the premise that writing of the entire character has been completed. The user input can not be reduced. In addition, this method includes a process for removing correspondence duplication in order to accurately recognize strokes with different stroke orders and this causes a great increase in processing amount.

  In view of the above problems, an object of the present invention is to provide a character search device capable of searching for characters or symbols by partial patterns even when a plurality of strokes are input in one stroke.

  In order to achieve the above object, a character search device according to the present invention compares an input pattern by a pointing device capable of handwriting input with a character or symbol pattern registered in a dictionary storage unit, and the input pattern. And a character search device that presents a character or symbol that matches or resembles the whole or part of the dictionary, and the dictionary storage unit is a dictionary that is a stroke constituting the character or symbol as data representing a character or symbol pattern Data representing a stroke and stroke order data representing the stroke order of each dictionary stroke are registered at least, and referring to the dictionary storage unit, an input stroke included in the input pattern and one dictionary stroke or a plurality of consecutive strokes Stroke similarity calculator for calculating similarity with a dictionary stroke sequence that is a dictionary stroke Based on the similarity obtained by the stroke similarity calculation unit, the combination of an input stroke included in one input pattern and a dictionary stroke sequence of one character or symbol registered in the dictionary storage unit is similar A similarity table creation unit that creates a degree table, and a dictionary stroke sequence that gives the maximum similarity to each input stroke is determined from the similarity table, and the sum of the maximum similarities for all input strokes is determined as the maximum similarity A similarity sum calculation unit obtained as a sum, a predetermined penalty condition for lowering the similarity between the input pattern and the character or symbol pattern registered in the dictionary storage unit with reference to the similarity table Is satisfied, a predetermined penalty value corresponding to the condition is subtracted from the maximum similarity sum to obtain a pattern matching degree, and a predetermined penalty condition is satisfied. A search result that presents characters or symbols as a search result in descending order of the pattern matching degree obtained by the penalty value subtracting unit using the maximum similarity sum as a pattern matching degree and the penalty value subtracting unit And a presentation unit.

  In the context of the present invention, “stroke” means a handwriting written from one pen-down to a pen-up. “Pattern” means a whole or partial pattern of characters or symbols formed from one or more “strokes”.

  According to the above configuration, the similarity table creating unit is configured to input the input stroke included in one input pattern based on the similarity obtained by the stroke similarity calculating unit and one registered in the dictionary storage unit. Create a similarity table for a combination of character or symbol dictionary stroke sequences, and the similarity sum calculation unit determines a dictionary stroke sequence that gives the maximum similarity for each input stroke from the similarity table, all By obtaining the sum of the maximum similarities with respect to the input strokes as the maximum similarity sum, even if a pattern that should originally be written in a plurality of strokes is input in one stroke, it is possible to search for characters or symbols by partial patterns. . The penalty value subtracting unit refers to the similarity table, and a predetermined penalty condition for lowering the similarity is established between the input pattern and the character or symbol pattern registered in the dictionary storage unit. If so, it is possible to improve the search accuracy of the character or symbol by the partial pattern by subtracting a predetermined penalty value corresponding to the condition from the maximum similarity sum to obtain the pattern matching degree.

  In the above character search device, the penalty condition is, for example, that a dictionary stroke sequence that gives the maximum similarity to one input stroke included in a certain input pattern corresponds to another input stroke included in the same input pattern. Even giving the maximum similarity.

  In the above character search device, the penalty condition includes, for example, a dictionary stroke corresponding to a certain input stroke with a similarity equal to or higher than a certain value, and the input stroke next to the input stroke after the dictionary stroke. In contrast, there is no corresponding dictionary stroke with a similarity greater than or equal to the predetermined value.

  In the above character search device, the penalty condition is, for example, that there is no dictionary stroke corresponding to the input stroke with a similarity greater than or equal to a certain value.

  In the above character search device, it is preferable that the stroke similarity calculation unit obtains the similarity between the input stroke and the dictionary stroke sequence according to an expansion / contraction matching method.

  In order to achieve the above object, a program according to the present invention compares an input pattern by a pointing device capable of handwriting input with a character or symbol pattern registered in a dictionary storage unit, and the input pattern And a program for causing a computer to execute a character search process that presents a character or symbol that is identical or similar in whole or part, and the dictionary storage unit stores the character or symbol as data representing a character or symbol pattern. Data representing a dictionary stroke, which is a constituent stroke, and stroke order data representing the stroke order of each dictionary stroke are registered, and the dictionary storage unit is referred to. Dictionary stroke sequence, which is a plurality of dictionary strokes A stroke similarity calculation process for determining the similarity of the input stroke, an input stroke included in one input pattern based on the similarity calculated by the stroke similarity calculation process, and one character registered in the dictionary storage unit Alternatively, a similarity table creation process for creating a similarity table for combinations of symbolic strokes with symbols, and a dictionary stroke sequence that gives the maximum similarity to each input stroke is determined from the similarity table, and all inputs are determined. Similarity sum calculation processing for obtaining the sum of the maximum similarities with respect to the stroke as the maximum similarity sum, between the input pattern and the character or symbol pattern registered in the dictionary storage unit with reference to the similarity table If a predetermined penalty condition for lowering the similarity is satisfied, a predetermined penalty value corresponding to the condition is calculated from the maximum similarity sum. The pattern matching degree is subtracted, and when a predetermined penalty condition is not satisfied, the penalty value subtraction process using the maximum similarity degree as the pattern matching degree, and the pattern matching degree obtained in the penalty value subtraction process in descending order. And a search result presentation process for presenting characters or symbols as search results.

  According to the present invention, it is possible to provide a character search device capable of searching for characters or symbols by partial patterns even when a plurality of images are input in one stroke.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  The present invention relates to an apparatus, method, program, etc. for searching for characters or symbols by partial patterns. The characters or symbols to be searched include arbitrary symbols in addition to kana, kanji, and numbers. included. In the present specification and claims, the search for these characters or symbols is abbreviated as “character search”.

  The present invention can be implemented as a character search program that is executed in a computer environment to which a pointing device capable of inputting handwriting such as a pen tablet or a mouse is connected. A recording medium on which the character search program is recorded is also an embodiment of the present invention. Note that the present invention may be implemented as a computer or a portable information terminal provided with an appropriate pointing device. Thus, when implementing the character search device of the present invention as a computer or a portable information terminal, the computer or the portable information terminal can have any function other than the character search function.

  Hereinafter, a computer in which the character search program according to the present invention is installed will be described as an embodiment of the character search device according to the present invention.

  FIG. 1 is a block diagram showing a functional configuration realized by executing the character search program of the present invention in the computer according to the present embodiment.

  As shown in FIG. 1, the computer 1 according to the present embodiment includes a pointing device 11 and a display 12 for allowing a user to input handwritten characters. Further, the computer 1 implements a character search function by using a stroke similarity calculation unit 13, a similarity table creation unit 14, a similarity sum calculation unit 15, a penalty value subtraction unit 16, and a stroke dictionary storage unit. 17, an input stroke storage unit 18, and a search result presentation unit 19. The stroke similarity calculating unit 13, the similarity table creating unit 14, the similarity sum calculating unit 15, the penalty value subtracting unit 16, and the search result presenting unit 19 are stored in a memory (not shown) of the processor (not shown) of the computer 1. It is functionally realized by executing a character search program installed in the system (not shown), and hardware corresponding to each of these units does not necessarily exist.

  As the pointing device 11, an input device capable of handwriting input such as a mouse or a pen tablet is used. When a display-integrated pen tablet is used as the pointing device 11, the display 12 can be omitted.

  When handwriting input is made by the user, the pointing device 11 acquires input stroke data for each stroke of the input pattern. The acquired input stroke data is stored in the input stroke storage unit 18. “One stroke” means a continuous trajectory drawn from pen down to pen up (in the case of a mouse, from drag start to drag end).

Here, the input stroke data is
(X _i , y _i , t _i ): 0 ≦ i <n
It is time-series feature point data expressed as follows. x _i and y _i are feature point coordinates, t _i is pen up / down information, i is a feature point number, and n is an integer. The feature point is a characteristic point extracted from the sampling data of the stroke. Various methods have been proposed for feature point extraction, such as a recursive farthest point separation method, a method of re-sampling at regular time intervals, and a method of quantizing a direction component. The present invention does not particularly depend on the feature point extraction method. When a plurality of strokes are input, the stroke data of those input strokes are given stroke numbers in the order of writing, and stored in the input stroke storage unit 18. In this embodiment, the stroke number is added. However, if the stroke data is saved in order, the stroke data string is saved, and the stroke order is saved, it is not always necessary to add the stroke number. Absent.

The stroke dictionary storage unit 17 stores dictionary stroke data in which various character or symbol patterns are represented by strokes constituting the pattern. Dictionary stroke data defined by time-series feature point data is stored. Dictionary stroke data is
(Vx _ij , vy _ij , vt _ij ): 0 ≦ i <m, 0 ≦ j <k
It is expressed. vx _ij and vy _ij are feature point coordinates, and vt _ij is pen-up / down information. j is a feature point number. i is a stroke number in the character pattern. The stroke number is given according to the stroke order of each stroke. m and k are integers.

  The stroke similarity calculation unit 13 refers to the stroke dictionary storage unit 17 and calculates the similarity between an input stroke and one dictionary stroke or a dictionary stroke sequence written continuously in the dictionary stroke group of one character. Ask. The obtained similarity is passed to the similarity table creation unit 14. Note that the “dictionary stroke sequence” means a plurality of strokes continuous in time series among a plurality of dictionary strokes constituting one character pattern. That is, when a user performs handwriting input, a portion that should be originally written with a plurality of strokes is often written with one stroke by continuous writing. For this reason, the stroke similarity calculation unit 13 obtains a similarity by sequentially comparing one input stroke with a plurality of dictionary strokes (that is, a dictionary stroke sequence).

  The stroke similarity calculation unit 13 uses an expansion / contraction matching method when obtaining the similarity between the input stroke and the dictionary stroke sequence. One of the most well-known methods for stretch matching is dynamic programming matching (so-called DP matching), but stretch matching applicable to the present invention is not limited to DP matching alone. In order to determine the number of strokes in the dictionary stroke sequence corresponding to one input stroke, the following method can be considered. For example, the similarity s (t) between one input stroke and a pattern formed from the stroke of the first stroke to the stroke of the t-th stroke in the dictionary stroke sequence is the same input stroke and 1 in the same dictionary stroke sequence. When the similarity s (t + 1) from the stroke of the stroke to the stroke of the (t + 1) stroke is larger, the stroke from the first stroke to the t-th stroke of the dictionary stroke sequence is greater than the one input stroke. It is a method of determining that it corresponds.

Thereby, for example, when there is an input stroke S _in1 as shown in FIG. 2A and a dictionary pattern as shown in FIG. It is possible to determine that the dictionary strokes corresponding to the input stroke S _in1 are S _dic1 and S _dic2 in FIG.

As another method for the stroke similarity calculation unit 13 to obtain the similarity between the input stroke and the dictionary stroke sequence, there is the following method. For example, the feature point (x _p, y _p) of the input stroke characteristic points of the dictionary stroke (vx _iq, vy _iq) represents the correspondence between [p, q] and its local evaluation value f (p, q ), The next correspondence [p ′, q ′] of the correspondence [p, q] is changed to f (p ′, q + 1) from [p + 1, q], [p, q + 1], [p + 1, q + 1]. Choose to maximize q ′) and repeat. Then, in the process of this iterative process, when both the input stroke and the dictionary stroke reach the final stroke, the matching is finished.

  Based on the processing result of the stroke similarity calculation unit 13, the similarity table creation unit 14 creates a dictionary stroke string similarity table for input strokes for combinations of input patterns and all character patterns registered in the dictionary ( Hereinafter, this table is referred to as a “similarity table”) and is stored at least temporarily. Note that “creating a similarity table” means that all combinations of input strokes and dictionary stroke sequences are determined for similarity, and does not necessarily require a tabular output.

Here, when creating the similarity table between one input pattern and one character pattern, the similarity table creating unit 14 creates a dictionary of each stroke constituting the one input pattern and the one character pattern. For the combination with the dictionary stroke sequence starting from each stroke, the similarity is calculated using the stroke similarity calculation unit 13. That is, if the number of strokes of the input pattern is n _in and the number of strokes of a certain character pattern registered _in the dictionary is n _dic , the combination of the input pattern and the character pattern has a size of n _in × n _dic . A similarity table will be created.

  For example, it is assumed that the number of strokes of the input pattern is 3, and the number of strokes of a certain character pattern registered in the dictionary is 5. In this case, with respect to the combination of the input pattern and the character pattern, the similarity between the three strokes of the following (1) to (5) is the stroke similarity for each of the three strokes of the input pattern. The 3 × 5 similarity table is created by the degree calculation unit 13. Each element of the similarity table is represented as ss (i, j). Here, i is the stroke number in the input pattern, and j is the dictionary stroke number that is the head of the dictionary stroke sequence.

(1) A dictionary stroke string composed of first to fifth strokes of dictionary strokes of character patterns;
(2) A dictionary stroke sequence comprising the second to fifth strokes,
(3) A dictionary stroke sequence consisting of the third to fifth strokes,
(4) A dictionary stroke sequence comprising the fourth and fifth strokes,
(5) A dictionary stroke sequence composed of the fifth stroke.

  The similarity sum calculation unit 15 selects a dictionary stroke sequence that gives the largest similarity to each input stroke from the similarity table created by the similarity table creation unit 14, and selects the maximum selected for all the input strokes. Find the sum st of similarities. That is, the sum st is obtained by the following equation.

st = Σ [i = 0 to (n _in −1)] max [ss (i, j) | 0 ≦ j <n _dic ]

  The penalty value subtracting unit 16 refers to the similarity table created by the similarity table creating unit 14, and calculates the similarity between the input pattern and the character or symbol pattern registered in the dictionary storage unit. When a predetermined penalty condition to be lowered is satisfied, a predetermined penalty value corresponding to the penalty condition is subtracted from the maximum similarity sum st calculated by the similarity sum calculation unit 15. The value obtained as a result is the pattern matching degree between the input pattern and the character pattern. The calculation of the penalty value will be described in detail later.

  The search result presentation unit 19 displays the character patterns on the display 12 as the search results in descending order of the pattern matching degree obtained for each combination of the input pattern and all the character patterns registered in the dictionary.

  Here, details of the processing of the penalty value subtraction unit 16 will be described. For example, the following first to third conditions can be considered as penalty conditions for the penalty value subtraction unit 16 to subtract penalty values.

The first condition is that the penalty value subtracting unit 16 refers to the similarity table created by the similarity table creating unit 14 and gives a dictionary stroke that gives the maximum similarity to one input stroke included in a certain input pattern. The column gives maximum similarity to another input stroke included in the same input pattern. In this case, the penalty value subtracting unit 16 considers that there is an overlap in the correspondence between the input stroke and the dictionary stroke, and subtracts the predetermined penalty value p ₁ from the maximum similarity sum st.

  As described above, in this embodiment, the stroke similarity calculation unit 13, the similarity table creation unit 14, and the similarity sum calculation unit 15 are similar by allowing overlapping of stroke correspondence between the input pattern and the character pattern in the dictionary. Processing to calculate the degree. However, as described above, if it is permitted that each stroke of the input pattern overlaps with a stroke other than the dictionary stroke that should correspond originally, for example, identification between characters having a pattern in which several similar strokes are arranged. Will become difficult. The penalty value subtraction according to the first condition serves to reduce the difficulty of identification due to overlapping strokes to some extent and increase the accuracy.

For example, the three input strokes S _{in1 to} S _in3 constituting the input pattern shown in FIG. _3A are compared with the dictionary strokes S _{dic1 to} S _{dic3 of the} character pattern shown in FIG. In this case, each of the input strokes S _in1 and S _in2 has the maximum similarity with respect to the dictionary stroke S _dic1 , and the input stroke S _in3 has the maximum similarity with respect to the dictionary stroke S _dic3 . If there is no penalty value subtraction by the penalty value subtraction unit 16, there are dictionary strokes having similar positions and shapes for all of the input strokes S _{in1 to} S _in3 , and therefore, the input pattern of FIG. The degree of pattern matching with the character pattern of (b) becomes very high. However, since the input pattern in FIG. 3A is not a partial pattern of the character pattern in FIG. 3B, it is not preferable that the pattern matching degree is too high. Therefore, the penalty value subtraction unit 16 can improve the accuracy of the search for the entire character by imposing a penalty on the overlapping strokes that give the maximum similarity when the first condition is satisfied. .

The second condition is that when there is a dictionary stroke A corresponding to a certain input stroke a with a similarity equal to or greater than a certain value, the dictionary stroke A is followed by a constant value for the input stroke b next to the input stroke a. There is no corresponding dictionary stroke with the above similarity. In this case, the penalty value subtraction unit 16 subtracts a predetermined penalty value p ₂ from the maximum similarity sum st. Thereby, a certain amount of penalty can be given to the stroke order difference.

  For example, the input pattern “white” in FIG. 4A is a partial pattern of both the dictionary pattern “clear” in FIG. 4B and the dictionary pattern “target” in FIG. It is assumed that the input pattern of FIG. 4A is input by handwriting in the correct stroke order of the character “white”. Here, for each stroke of the input pattern shown in FIG. 4A, the dictionary pattern “Fine” shown in FIG. 4B also has a very similar position as the dictionary pattern “Target” shown in FIG. 4C. And a dictionary stroke with a shape. Therefore, if only the similarity between the input stroke and the dictionary stroke sequence is taken into consideration, the dictionary pattern “Fine” in FIG. 4B is different from the input pattern “White” in FIG. ) In the same row as the dictionary pattern “target”, the higher the degree of similarity. However, for the input pattern of FIG. 4A, it is desirable that the character having “white” on the left side is higher in the search results than the character having “fish” on the left side.

Each stroke of the dictionary pattern “target” in FIG. 4C corresponds to the input pattern in FIG. 4A in order with high similarity. However, each stroke of the dictionary pattern “Fine” in FIG. 4B does not necessarily correspond with high similarity to the input pattern in FIG. For example, the first stroke S _in1 of the input pattern “white” in FIG. 4A and the first stroke S _dic1 of the dictionary pattern “ _Bright ” in FIG. 4B are similar in both position and shape. And high similarity can be taken. However, the second stroke S _in2 of the input pattern “white” in FIG. _4A and the second stroke S _dic2 of the dictionary pattern “fresh” in FIG. Is low.

Therefore, according to the second condition, it is determined whether the dictionary stroke is associated with a high degree of similarity according to the writing order of the input strokes. If not, the predetermined penalty value p _{2 is set} to the maximum similarity sum. Subtract from st. As a result, the dictionary pattern “Target” in FIG. 4C is more in the search result than the dictionary pattern “Fine” in FIG. 4B with respect to the input pattern “White” in FIG. The effect that the ranking of becomes higher is obtained.

The third condition is that there is an input stroke that does not correspond to a dictionary stroke with a certain degree of similarity or more, in other words, there is no dictionary stroke that can be regarded as corresponding to a certain input stroke. At this time, the penalty value subtraction unit 16 subtracts a predetermined penalty value p ₃ from the maximum similarity sum st.

For example, the strokes S _{in1 to} S _in5 of the input pattern in FIG. 5A correspond to the strokes S _{dic1 to} S _dic5 of the dictionary pattern “dragon” in FIG. _5B with very high similarity. However, the stroke S _in6 of the input pattern in FIG. _5A is not similar to any part of the dictionary pattern “dragon” in FIG. Considering only the similarity between the input stroke and the dictionary stroke sequence, the dictionary pattern “dragon” in FIG. 5B has a stroke similar to the input stroke S _in6 compared to the input pattern in FIG. Although it is not included, it has a high pattern matching degree like the dictionary pattern “part” in FIG. 5C including a stroke similar to the input stroke _Sin6 .

Therefore, according to the third condition, when a part of the input pattern is not similar to any part of the dictionary pattern, the predetermined penalty value p _{3 is set} to the maximum similarity so that the characters of the dictionary pattern do not become higher in the search result. Subtract from the sum st.

Incidentally, the penalty value p ₁ ~p ₃ is applied according to each of the first to third conditions may be determined empirically appropriate value. For example, when the pattern matching degree obtained as a calculation result of the penalty value subtracting unit 16 can take a value of 0 to 1000, the penalty value is set in increments of a predetermined value (for example, 50 by increments) while performing simulation using various handwritten character patterns. ) And adopt a penalty value that gives the most appropriate search result.

  As described above, the first to third specific conditions are exemplified for the penalty value application condition by the penalty value subtracting unit 16. However, when the penalty value is calculated by the penalty value subtracting unit 16, It is also effective to apply a combination of the three conditions.

  Next, the procedure of the character search method according to the present invention will be described with reference to FIG.

  When the user inputs a handwritten pattern using the pointing device 11, the pointing device 11 acquires input stroke data for each stroke of the input pattern and stores it in the input stroke storage unit 18 (step Op1).

  Next, the similarity table creation unit 14 selects the first stroke of the input pattern and the first character pattern registered in the stroke dictionary storage unit 17, and sets the input stroke and the dictionary stroke sequence of the character pattern. A similarity table is created (Step Op2). In this step Op2, when the similarity between the input stroke and the dictionary stroke string is obtained, the stroke similarity calculation unit 13 uses, for example, an expansion / contraction matching technique.

The similarity table creation unit 14 makes all combinations of input strokes and dictionary stroke sequences until the similarity table is completed, that is, if the number of input strokes is n _in and the number of dictionary strokes of the character pattern is n _dic. Step 2 is repeatedly executed until a similarity table having a size of n _in × n _dic is completed (Yes in Step Op3).

  Next, the similarity sum calculation unit 15 selects a dictionary stroke sequence that gives the largest similarity to each input stroke from the similarity table created in Step Op2, and selects the maximum similarity selected for all the input strokes. The sum of degrees st is obtained (step Op4).

  Next, the penalty value subtracting unit 16 calculates a penalty value based on the similarity table created in Step Op2, and subtracts the calculated penalty value from the maximum similarity sum st calculated in Step Op4. Then, the pattern matching degree between the input pattern and the character pattern is obtained (Step Op5).

  The above Steps Op2 to Op5 are repeated until the pattern matching degree is obtained for the combination of the input pattern and all the character patterns registered in the stroke dictionary storage unit 17 (Yes in Step Op6).

  Finally, the search result presentation unit 19 displays character patterns as search results on the display 12 in descending order of pattern matching with respect to the input pattern (Op 7).

  With the above procedure, even when a plurality of images are input with one stroke, it is possible to efficiently and accurately search for a character including a pattern that partially matches the input pattern.

  INDUSTRIAL APPLICABILITY The present invention can be used as a character search device capable of searching for characters using partial patterns even when a plurality of images are input with one stroke, and a computer program that embodies this.

It is a block diagram of the character search device concerning one Embodiment of this invention. It is a figure for demonstrating matching with the input pattern (a) and dictionary pattern (b) in the character search device concerning one Embodiment of this invention. It is the figure which showed an example of the combination of the input pattern (a) and dictionary pattern (b) in case the penalty condition is imposed in the character search device concerning one Embodiment of this invention. No penalty condition is imposed on an example of a combination of an input pattern (a) and a dictionary pattern (b) when a penalty condition is imposed in the character search device according to an embodiment of the present invention, and a combination of the input pattern. It is the figure which showed the example of the dictionary pattern (c). No penalty condition is imposed on an example of a combination of an input pattern (a) and a dictionary pattern (b) when a penalty condition is imposed in the character search device according to an embodiment of the present invention, and a combination of the input pattern. It is the figure which showed the example of the dictionary pattern (c). It is a flowchart which shows an example of the procedure of the character search method concerning this invention. It is an example of the screen of the conventional partial pattern character search device.

Explanation of symbols

1 Computer (character search device)
DESCRIPTION OF SYMBOLS 11 Pointing device 12 Display 13 Stroke similarity calculation part 14 Similarity table preparation part 15 Similarity sum calculation part 16 Penalty value subtraction part 17 Stroke dictionary memory | storage part 18 Input stroke memory | storage part 19 Search result presentation part

Claims (6)

A character search device that compares an input pattern by a pointing device capable of handwriting input with a character or symbol pattern registered in a dictionary storage unit and presents a character or symbol that matches or resembles the whole or part of the input pattern. Because
In the dictionary storage unit, as data representing a character or symbol pattern, at least data representing a dictionary stroke which is a stroke constituting the character or symbol and stroke order data representing the stroke order of each dictionary stroke are registered,
A stroke similarity calculation unit that refers to the dictionary storage unit and obtains a similarity between an input stroke included in the input pattern and a dictionary stroke sequence that is one dictionary stroke or a plurality of dictionary strokes written continuously,
Based on the similarity obtained by the stroke similarity calculation unit, the similarity for a combination of an input stroke included in one input pattern and a dictionary stroke sequence of one character or symbol registered in the dictionary storage unit A similarity table creation section for creating a table;
A similarity sum calculation unit that determines a dictionary stroke sequence that gives the maximum similarity for each input stroke from the similarity table, and calculates the sum of the maximum similarities for all input strokes as the maximum similarity sum;
Referring to the similarity table, if a predetermined penalty condition for lowering the similarity is satisfied between the input pattern and the character or symbol pattern registered in the dictionary storage unit, the condition is A penalty value subtracting unit that subtracts a corresponding predetermined penalty value from the maximum similarity sum to obtain a pattern matching degree, and when the predetermined penalty condition is not established, the maximum similarity sum to be a pattern matching degree;
A character search apparatus comprising: a search result presenting unit that presents characters or symbols as a search result in descending order of the degree of pattern matching obtained by the penalty value subtracting unit.   The dictionary stroke sequence in which the penalty condition gives the maximum similarity to one input stroke included in a certain input pattern gives the maximum similarity to another input stroke included in the same input pattern The character search device according to claim 1, wherein   The penalty condition has a dictionary stroke corresponding to a certain input stroke with a certain degree of similarity, and is similar to the input stroke following the input stroke after the dictionary stroke with the certain value or more. The character search device according to claim 1, wherein there is no corresponding dictionary stroke in degrees.   The character search device according to claim 1, wherein the penalty condition is that there is no dictionary stroke corresponding to the input stroke with a similarity greater than or equal to a certain value.   The character search device according to any one of claims 1 to 4, wherein the stroke similarity calculation unit obtains a similarity between an input stroke and a dictionary stroke sequence according to an expansion / contraction matching method. Character search processing that compares an input pattern by a pointing device capable of handwriting input with a character or symbol pattern registered in the dictionary storage unit and presents a character or symbol that matches or resembles the whole or part of the input pattern A program for causing a computer to execute
In the dictionary storage unit, as data representing a character or symbol pattern, at least data representing a dictionary stroke which is a stroke constituting the character or symbol and stroke order data representing the stroke order of each dictionary stroke are registered,
A stroke similarity calculation process for obtaining a similarity between an input stroke included in the input pattern and a dictionary stroke sequence that is a plurality of dictionary strokes written continuously, with reference to the dictionary storage unit;
Based on the similarity obtained by the stroke similarity calculation process, the similarity for a combination of an input stroke included in one input pattern and a dictionary stroke sequence of one character or symbol registered in the dictionary storage unit Similarity table creation processing to create a table,
A similarity sum calculation process for determining a dictionary stroke sequence that gives the maximum similarity for each input stroke from the similarity table, and calculating a sum of the maximum similarities for all input strokes as a maximum similarity sum;
Referring to the similarity table, if a predetermined penalty condition for lowering the similarity is satisfied between the input pattern and the character or symbol pattern registered in the dictionary storage unit, the condition is A penalty value subtraction process in which a predetermined penalty value is subtracted from the maximum similarity sum to obtain a pattern matching degree, and when the predetermined penalty condition is not satisfied, the maximum similarity sum is used as a pattern matching degree;
A program that causes a computer to execute search result presentation processing that presents characters or symbols as search results in descending order of the degree of pattern matching obtained in the penalty value subtraction processing.

Images