JP2007026027A - Character recognition program, character recognition device and character recognition method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the reliability of character recognition by selecting the optimal selection system according to the degree of deterioration of characters, and to quickly perform character recognition without requiring a processing time. <P>SOLUTION: This character recognition program for making a computer execute character recognition is provided to make the computer execute a deterioration degree estimation step for dividing the density images of segmented character regions into at least three concentration regions, and for estimating the degree of the deterioration of characters as the degrees of deterioration based on the density regions, a character recognition system selection step for selecting at least one character recognition system from the plurality of character recognition systems based on the degree of deterioration estimated by the deterioration degree estimation step and a recognition step for recognizing the characters based on the recognition result by the character recognition system selected by the character recognition system selection step. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a character recognition program, a character recognition device, and a character recognition method for recognizing characters from a cut character image.

  In recent years, with the spread of digital video, digital cameras, and camera-equipped mobile phones, the number of images input to the cameras has increased. If characters included in these images can be recognized, information retrieval using the characters in the image as a trigger can be realized.

  However, while the character images input by the scanner are generally of high quality, the images input by the camera have various degrees of deterioration from high-quality images to low-quality images depending on the distance to the subject and the degree of blurring. Contains characters. Conventionally, several technical documents are known as recognition methods for characters with various degrees of degradation (see, for example, Patent Documents 1 to 5).

  In Patent Document 1, binarization is performed on a grayscale image by changing a threshold value, and a plurality of dictionaries having different levels of collapse are created in advance. At the time of recognition, an optimal dictionary is selected based on the estimated deterioration of input characters.

  In Patent Document 2, a plurality of dictionaries with different character qualities are prepared, and the dictionaries are used properly according to the quality of the input document.

  In Patent Literature 3, a standard dictionary and an extended dictionary are created, and characters that cannot be recognized by the standard dictionary are tried to be recognized by the extended dictionary. Specifically, the recognition is first performed with the standard dictionary, and the recognition with the extended dictionary is performed only when the recognition reliability is low.

  In Patent Document 4, a character deterioration degree is estimated for a binary pattern, and character characteristics are corrected based on the estimated character deterioration degree.

In Patent Document 5, a binary recognition system and a light and shade recognition system are provided, and when the former fails, the latter performs recognition.
JP-A-9-343292 JP-A-8-241378 JP 2000-82113 A JP 2003-256774 A JP-A-11-66240

  When characters input from a camera are compared with characters input by a scanner, the resolution is generally low and the quality is low, and a conventional OCR (Optical Character Reader) cannot always be recognized accurately. Therefore, several character recognition methods suitable for low-quality images have been proposed, but conversely, these methods do not allow recognition accuracy as high as that of conventional OCR for high-quality images. . Although a method for fusing these has been proposed, it has not been a method capable of accurately recognizing high-quality images to low-quality images.

  In Patent Document 1, a dictionary is selected based on input character deterioration estimation. However, since the ratio of black pixels between the input character and the reference pattern is used in the deterioration estimation, there are a plurality of fonts. In some situations, it is difficult to accurately estimate deterioration.

  In Patent Document 2 and Patent Document 3, a plurality of dictionaries having different character qualities are prepared and used properly, but the quality and the character recognition method are not associated with each other, and the character recognition method suitable for high-quality characters is low quality. Characters are not always optimal, so there are cases where it is not possible to deal with them by using different dictionaries.

  In Patent Document 4, a character deterioration degree is estimated for a binary pattern, and character characteristics are corrected based on the estimated character deterioration degree. However, in the case of a low-quality character having a small size, there is a problem that character structure information is easily lost by binarization.

  Patent Document 5 has a binary-based recognition system and a grayscale image-based recognition system, and when the former fails, the latter performs recognition. If the entire document is of low quality, the recognition accuracy will be higher if the grayscale image-based recognition system is used from the beginning, but this method, which always recognizes and verifies with the binary-based recognition system, takes time. There is.

  According to the present invention, it is possible to select an optimum recognition method corresponding to the degree of deterioration of characters, to increase the reliability of character recognition, and to perform character recognition quickly without requiring processing time. An object of the present invention is to provide a character recognition program, a character recognition device, and a character recognition method.

  In order to solve the above-described problems, the present invention is a character recognition program for causing a computer to perform character recognition, dividing a grayscale image of a cutout character region into at least three density regions, and character based on these density regions. A deterioration degree estimation step for estimating the degree of deterioration of the character as a deterioration degree, and a character recognition method selection for selecting at least one character recognition method from a plurality of character recognition methods based on the deterioration degree estimated by the deterioration degree estimation step Causing the computer to execute a step and a recognition step of recognizing a character based on a recognition result by the character recognition method selected in the character recognition method selection step.

  In this character recognition program, in the character recognition method selection step, one of a method using a character direction feature and a method using a character pixel feature is selected. In the deterioration level estimation step, the grayscale image is divided into three density regions of white, black, and intermediate colors, and the character deterioration level is estimated based on the ratio of the intermediate color image region to the entire image region. .

  In addition, the present invention is a character recognition device that performs character recognition, and divides a gray-scale image of a cut-out character region into at least three density regions and estimates a degree of character degradation based on these density regions. An estimation unit, a character recognition method selection unit that selects at least one character recognition method from a plurality of character recognition methods based on the deterioration degree estimated by the deterioration degree estimation unit, and the character recognition method selection unit. A recognition unit for recognizing a character based on a recognition result obtained by the character recognition method.

  Note that each character has a storage unit that stores a recognition rate of each character recognition method with respect to the degree of deterioration of the character, the character recognition method selection unit, based on the degree of deterioration estimated by the deterioration degree estimation unit, The character recognition method can be selected by referring to the recognition rate stored in the storage unit.

  The present invention is also a character recognition method for causing a character recognition device to perform character recognition, wherein a grayscale image of a cut out character region is divided into at least three density regions, and the degree of character degradation based on these density regions A degradation level estimation step for estimating a degradation level, a character recognition mode selection step for selecting at least one character recognition mode from a plurality of character recognition modes based on the degradation level estimated by the degradation level estimation step, and A recognition step for recognizing a character based on a recognition result obtained by the character recognition method selected in the character recognition method selection step.

  As described in detail above, according to the present invention, it is possible to select an optimum recognition method corresponding to the degree of deterioration of characters, increase the reliability of character recognition, and require processing time. Character recognition can be performed quickly.

  Embodiments of the present invention will be described below with reference to the drawings. Note that the character recognition device according to the present embodiment can be used in a personal computer (PC), a personal digital assistant (PDA), a mobile phone, or the like with a built-in character recognition function. Further, not only character images (electrical signals) photographed by a camera or the like but also character images (electrical signals) by FAX or the like are targeted.

  FIG. 1 is a block diagram showing a configuration of a character recognition apparatus according to an embodiment of the present invention.

  The character recognition device shown in FIG. 1 divides a cut-out character region 1 from a grayscale image including characters and the cut-out character region into at least three density regions, and determines the degree of character degradation based on these density regions. Degradation level estimation unit 2 to be estimated, character recognition mode selection unit 3 that selects at least one character recognition mode from a plurality of character recognition modes based on the degradation level estimated by the degradation level estimation unit 2, and a character recognition mode In addition to performing character recognition using the character recognition method selected by the selection unit 3, a recognition unit 4 that recognizes and outputs a final character based on the recognition result is provided.

  In the present embodiment, the recognition unit 4 includes a method-compatible character recognition unit 5 and an integration unit 6 that integrates the recognition results.

  The system-compatible character recognition unit 5 divides the system roughly into a first character recognition system (binary character recognition system) 5A including one or a plurality of different systems that use character direction features, and character pixel characteristics ( The second character recognition method (degraded character recognition system) 5B including one or a plurality of different methods using density characteristics) has a function of recognizing each character. These first and second character recognition methods Any one of the methods is selected by the character recognition method selection unit 3 and functions.

  The characteristics of each of the above systems will be briefly described. For example, the robustness of character recognition with respect to various fonts is higher in the system using the direction feature (first character recognition system), and the character characteristics deterioration due to blurring is the density characteristics. The method using the (second character recognition method) is generally considered to be higher. Therefore, in the present embodiment, a method (second character recognition system) that uses density features is adopted if the degree of deterioration is greater than a predetermined threshold with respect to the degree of deterioration calculated for the input character. If is small, a method using a directional feature (first character recognition system) is adopted. In each recognition system, a plurality of different methods belonging to the recognition system can be prepared. In the following, an example in which the binary character recognition system and the deteriorated character recognition are used properly according to the degree of deterioration is shown.

  The integration unit 6 outputs one character recognition result from the recognition result recognized by the method-compatible character recognition unit 5, as will be described later.

  The operation of the embodiment of the present invention will be described below using the flowchart of FIG.

  First, when performing character recognition, the cutout unit 1 cuts out characters from the grayscale image having the input character string (step S1), and the cutout characters are input (step S2). This cut-out uses a known method conventionally used in character recognition, and a description thereof is omitted here.

  Next, the degradation level estimation unit 2 estimates the degradation level of characters (step S3). The deterioration degree estimation operation will be described in detail with reference to the flowchart of FIG.

  First, size normalization is performed on the input shaded character image (0-255), and the number of pixels is set to N × N (step ST1). Next, after performing canonicalization, the norm is set to 1 (step ST2). Next, density conversion is linearly performed so that each pixel value falls within the range of 0 to 255 (step ST3). Then, the number of pixels having the value of the gray component (1-254) excluding white and black is counted (step ST4), and the degree of deterioration defined by the following equation is obtained (step ST5).

    Degree of degradation = 100 × number of gray components / (N × N)

  The canonicalization means an operation of leaving only a component orthogonal to a predetermined feature vector for a certain feature vector. When character shading itself is a feature, the black pixel component in the character image increases as the degree of character crushing increases. And as the degree of crushing increases, it is expected to eventually become full black pixels. Therefore, an operation is performed to leave only the components orthogonal to the characteristics of the entire black pixel (see Taizo Iijima, “Pattern Recognition Theory,” Morikita Publishing, 1989).

  When the degree of deterioration is estimated in this way, it is then determined whether or not the degree of deterioration estimated by the character recognition method selection unit is equal to or greater than a threshold value (step S4). 3 selects a binary character recognition system that is the first character recognition method 5A in the method-compatible character recognition unit 5 (step S5). When character recognition is performed in the first character recognition method 5A, the integration unit 6 next determines whether there are a plurality of recognition methods in the first character recognition method 5A (step S6). (Step S6, Y), the recognition results are integrated (Step S7).

  In the integration of recognition results (step S7), as shown in FIG. 4, the frequency of recognition results is counted for each recognized character (for each category), and the character having the maximum frequency is output as a recognized character. . If it is determined in step S6 that there is one recognition method in the first character recognition method 5A (step S6, N), the character recognized by that method is output as a recognized character.

  On the other hand, if it is determined in step S4 that the degree of deterioration is equal to or greater than the threshold (step S4, Y), the character recognition method selection unit is 3, and the deterioration corresponding to the second character recognition method 5B in the method-compatible character recognition unit 5 is performed. A character recognition system is selected (step S8). When character recognition is performed in the second character recognition method 5B, the integration unit 6 next determines whether there are a plurality of recognition methods in the second character recognition method 5B (step S9). (Step S9, Y), the above recognition results are integrated (Step S7).

  The operation of the integration unit 6 is the same as that when the first character recognition method (binary character recognition system) 5A is selected, and the description of the following operation is omitted.

  As the first character recognition method (binary character recognition system) 5A, the following method is known and used in the present embodiment.

(1) Sonning, Toru Tahara, Hiroki Aki, Masayuki Kimura,
`` High-precision character recognition using directional element features, ''
Theory of Science (D-II) vol.J74-D-II no.3, pp.330-339, Mar. 1991.
(2) Shinji Tsuruoka, Masanori Kurita, Tomio Harada, Fumitaka Kimura, Koji Miyake,
`` Handwritten kanji and hiragana recognition by weighted direction index histogram method, ''
Theory of Science (D) vol.J70-D no.7, pp.1390-1397, July 1987.
(3) Character recognition method based on outline direction contribution feature Hirota et al., "Identification of handwritten Chinese characters by outline direction contribution feature"
IEICE Transactions '83 / 10 Vol.J66-D No.10, pp.1185-1192
(4) Character recognition method by multi-component compression method ▲ 裴 ▼ and others, "One handwritten Kanji character recognition method-Recognition by multi-component compression method and partial pattern method-"
IEICE Transactions '85 / 4 Vol.J68-D No.4, pp.773-780
(5) Character recognition method by direction pattern matching method Saito et al., "Analysis by hand pattern kanji direction pattern matching method"
IEICE Transactions '82 / 5 Vol.J65-D No.5, pp.550-557

  In the second character recognition method (degraded character recognition system) 5B, the following method is known.

J. Sun, Y. Hotta, Y. Katsuyama and S. Naoi, "Low Resolution Character Recognition by Dual Eigenspace and Synthetic Degraded Patterns," Proc. Of the 1 ^st ACM Workshop on Hardcopy Document Processing, pp.15-22, 2004.

Embodiment 2. FIG.

The second embodiment of the present invention will be described below.
In Embodiment 1, the case where the two recognition methods are selectively used according to the degree of deterioration has been described. In the second embodiment, a case will be described in which a character recognition method having the highest recognition rate corresponding to the degree of character deterioration is directly selected.

  FIG. 5 is a block diagram showing a configuration of the second embodiment. 5, the same reference numerals as those in FIG. 1 indicate the same or equivalent parts, and the description thereof is omitted here. In the second embodiment, the character recognition method selection unit 3A includes a deterioration degree / recognition rate correspondence table 11 in which a deterioration degree and a recognition rate are associated with each other as shown in FIG. Based on this table 11, a character recognition method used for character recognition is selected. As described in the first embodiment, the character recognition method may not necessarily be divided into the first character recognition method 5A and the second character recognition method 5B.

  FIG. 7 is an explanatory diagram conceptually showing the creation operation of this table. Learning data is prepared from those having a low degree of deterioration (no deterioration) to those having a high degree of deterioration, and all the tables provided with the learning data are prepared. The recognition rate is recognized by the recognition method, the recognition rate is determined from the correct answer rate, and the recognition rate is stored in a table.

The operation of the second embodiment will be described below using the flowchart of FIG.
First, the operation from step S1 to step S3 is the same as the operation from step S1 to step S3 in the first embodiment. Next, in step S14, the character recognition method selection unit 3A searches the deterioration degree / recognition rate correspondence table 11 based on the deterioration degree obtained in step S3, and has the highest recognition rate in the obtained deterioration degree. Select a high character recognition method. The recognition unit 4 performs character recognition using the selected character recognition method by the method-compatible character recognition unit 5 and causes the integration unit 6 to output the recognition result.

  In this case, the integration unit 6 performs the implementation when a plurality of character recognition methods (regardless of the first character recognition method or the second character recognition method) are selected by the character recognition method selection unit 3A for the degree of deterioration. The frequency of the recognition result is counted for each recognized character (for each category) as in the first mode, and the character having the maximum frequency is output as a recognized character. Of course, if there is only one selected method, the recognition result by that method is output as a recognized character as it is.

  According to the embodiment of the present invention, the following effects can be obtained.

  In the embodiment of the present invention, the recognition method is selectively used based on the deterioration degree of the input character. Therefore, it is possible to estimate the deterioration degree directly from the grayscale image as compared with the method of calculating the deterioration degree from the binary image. Yes, it is not affected by binarization. In addition, compared with a method of using a dictionary for each degree of deterioration, according to the embodiment of the present invention, a recognition method is used for each degree of deterioration, so that more accurate recognition can be realized. In addition, since the character recognition method is selectively used according to the degree of deterioration, it can be recognized at higher speed than the method of sequentially recognizing with a plurality of recognition methods.

  As mentioned above, although embodiment of this invention was described about the 1st, 2nd character recognition system, it cannot be overemphasized that this invention is applicable also to character recognition systems other than these character recognition systems.

  In addition, according to the embodiment of the present invention, a program for causing a computer constituting the character recognition apparatus to execute the steps described above with reference to FIGS. 2 and 8 can be provided as a character recognition program. By storing the above-described program in a computer-readable recording medium, the computer constituting the language analysis apparatus can be executed. Here, as the recording medium readable by the computer, a portable storage medium such as a CD-ROM, a flexible disk, a DVD disk, a magneto-optical disk, an IC card, a database holding a computer program, or another computer In addition, the database and the transmission medium on the line are also included.

(Appendix 1)
A character recognition program for causing a computer to perform character recognition,
A degradation level estimation step of dividing the grayscale image of the clipped character region into at least three density regions and estimating the degree of degradation of the character as a degradation level based on these density regions;
A character recognition method selection step of selecting at least one character recognition method from a plurality of character recognition methods based on the deterioration level estimated by the deterioration level estimation step;
A character recognition program for causing a computer to execute a recognition step for recognizing a character based on a recognition result obtained by the character recognition method selected in the character recognition method selection step.
(Appendix 2)
In the character recognition program described in Appendix 1,
In the character recognition method selection step, a character recognition program that selects one of a method using a character direction feature and a method using a character pixel feature.
(Appendix 3)
In the character recognition program described in Appendix 1,
When there are a plurality of character recognition methods for the selected character recognition method, at least one character recognition method is further selected from these character recognition methods based on the degree of deterioration.
(Appendix 4)
When there are a plurality of character recognition methods for the selected character recognition method, the character recognition step performs character recognition by all these character recognition methods, and recognizes characters based on these recognition results. A character recognition program.
(Appendix 5)
In the character recognition program according to any one of supplementary notes 1 to 4,
In the deterioration level estimation step, the grayscale image is divided into three density regions of white, black, and intermediate colors, and character deterioration is estimated based on the ratio of the intermediate color image region to the entire image region. program.
(Appendix 6)
A character recognition device for character recognition,
A degradation degree estimation unit that divides the grayscale image of the clipped character region into at least three density regions and estimates the degree of character degradation based on these density regions;
A character recognition method selection unit that selects at least one character recognition method from a plurality of character recognition methods based on the deterioration degree estimated by the deterioration degree estimation unit;
A character recognition device comprising: a recognition unit that recognizes a character based on a recognition result by the character recognition method selected by the character recognition method selection unit.
(Appendix 7)
It has a storage unit that stores the recognition rate by each character recognition method with respect to the deterioration degree of the character for each character,
The character recognition method selection unit selects the character recognition method by referring to a recognition rate stored in the storage unit based on the degree of deterioration estimated by the deterioration degree estimation unit. Character recognition device.
(Appendix 8)
A character recognition method for causing a character recognition device to perform character recognition,
A degradation level estimation step of dividing the grayscale image of the clipped character region into at least three density regions and estimating the degree of degradation of the character as a degradation level based on these density regions;
A character recognition method selection step of selecting at least one character recognition method from a plurality of character recognition methods based on the deterioration level estimated by the deterioration level estimation step;
A character recognition method comprising: recognizing a character based on a recognition result obtained by the character recognition method selected in the character recognition method selection step.

It is a block diagram which shows the structure of Embodiment 1 of this invention. It is a flowchart which shows the operation | movement of Embodiment 1 of this invention. It is a flowchart which shows the operation | movement of deterioration degree estimation. It is explanatory drawing which shows operation | movement of an integration part. It is a block diagram which shows the structure of Embodiment 2 of this invention. It is a deterioration degree / recognition rate correspondence table. It is explanatory drawing which shows the operation | movement at the time of preparation of a deterioration degree / recognition rate correspondence table. It is a flowchart which shows the operation | movement of Embodiment 2 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Cutout part, 2 Deterioration degree estimation part, 3, 3A Character recognition system selection part, 4 Recognition part, 5 System corresponding | compatible character recognition part, 6 Integration part (output part), 11 Deterioration degree / recognition rate correspondence table

Claims (5)

A character recognition program for causing a computer to perform character recognition,
A degradation level estimation step of dividing the grayscale image of the clipped character region into at least three density regions and estimating the degree of degradation of the character as a degradation level based on these density regions;
A character recognition method selection step of selecting at least one character recognition method from a plurality of character recognition methods based on the deterioration level estimated by the deterioration level estimation step;
A character recognition program for causing a computer to execute a recognition step for recognizing a character based on a recognition result obtained by the character recognition method selected in the character recognition method selection step. The character recognition program according to claim 1,
In the character recognition method selection step, a character recognition program that selects either one of a method using a character direction feature and a method using a character pixel feature. In the character recognition program according to claim 1 or 2,
In the deterioration level estimation step, the grayscale image is divided into three density regions of white, black, and intermediate colors, and character deterioration is estimated based on the ratio of the intermediate color image region to the entire image region. program. A character recognition device for character recognition,
A degradation degree estimation unit that divides the grayscale image of the clipped character region into at least three density regions and estimates the degree of character degradation based on these density regions;
A character recognition method selection unit that selects at least one character recognition method from a plurality of character recognition methods based on the deterioration degree estimated by the deterioration degree estimation unit;
A character recognition device comprising: a recognition unit that recognizes a character based on a recognition result by the character recognition method selected by the character recognition method selection unit. A character recognition method for causing a character recognition device to perform character recognition,
A degradation level estimation step of dividing the grayscale image of the clipped character region into at least three density regions and estimating the degree of degradation of the character as a degradation level based on these density regions;
A character recognition method selection step of selecting at least one character recognition method from a plurality of character recognition methods based on the deterioration level estimated by the deterioration level estimation step;
A character recognition method comprising: recognizing a character based on a recognition result obtained by the character recognition method selected in the character recognition method selection step.

Images