JP4869365B2 - Image processing apparatus and image processing method - Google Patents

Description

  The present invention relates to an image processing apparatus and an image processing method.

  An example of the image processing apparatus is a character recognition apparatus that recognizes characters included in an image. In a character recognition device, a recognition technique called a mutual subspace method has been proposed as one technique for character recognition using a plurality of feature quantities. This recognition technique recognizes by paying attention to various features of one character image (see Non-Patent Document 1, for example).

  On the other hand, a character image has a variation factor due to a positional deviation, an angle, and the like for each character image. In response to this, a technique called a subspace method has been disclosed as one method for performing pattern recognition, but the corresponding capability is not necessarily perfect (see, for example, Non-Patent Document 2).

Kenichi Maeda, Sadaichi Watanabe “Pattern Matching Method Introducing Local Structure”, IEICE Transactions Vol.J68-D, No.3, 1985. Kenichiro Ishii et al. “Easy-to-understand pattern recognition” (1998), Ohm

  In the subspace method, for example, if a large number of these fluctuation factors are not included in the learning pattern, it may not necessarily exhibit a large capacity in practice. Also, if the variation of the input image is too large, even the subspace method cannot be handled, and the recognition accuracy is reduced.

  SUMMARY An advantage of some aspects of the invention is that it provides an image processing apparatus and an image processing method capable of improving the accuracy of recognizing characters from a character image.

In order to solve the above-described problems, an image processing apparatus according to the present invention includes a memory in which a document image is stored, a recognition dictionary in which characters and feature data are stored correspondingly, and a document image read from the memory. A preprocessing unit that performs predetermined preprocessing on the character image to generate a character image , and a first image processing unit that performs a first image processing process on the character image generated by the preprocessing unit and has a plurality of different character patterns A first pattern generation unit that generates a character pattern group, and a second character pattern group having a plurality of different character patterns by performing a second image processing process on the character image generated by the preprocessing unit A second pattern generation unit; a first feature extraction unit that extracts feature data from each of a plurality of character patterns of the first character pattern group generated by the first pattern generation unit; and the second pattern generation unit. A second feature extraction unit for extracting feature data from each of a plurality of character patterns of the second character pattern group generated by the character generation unit, a plurality of feature data extracted by the first feature extraction unit, and the recognition A first similarity calculation unit that calculates the similarity between the character feature data stored in the dictionary, a plurality of feature data extracted by the second feature extraction unit, and a character stored in the recognition dictionary A second similarity calculating unit for calculating a similarity with the feature data, a similarity calculated by the first similarity calculating unit, and a similarity calculated by the second similarity calculating unit are predetermined. And a similarity integration unit for selecting characters having high similarity from the recognition dictionary using the integrated similarity.

The image processing method of the present invention includes a memory in which a document image is stored, a recognition dictionary in which characters and their feature data are stored correspondingly, a preprocessing unit, a first pattern generation unit, a second pattern generation unit, and a first feature. In an image processing method by an image processing apparatus having an extraction unit, a second feature extraction unit, a first similarity calculation unit, a second similarity calculation unit , and a similarity integration unit, the preprocessing unit reads a document image from the memory A step of performing a predetermined pre-processing on the read document image to generate a character image , and a plurality of different characters by the first pattern generation unit performing a first image processing process on the character image . Generating a first character pattern group having a pattern, and a second character pattern having a plurality of different character patterns by performing a second image processing process on the character image by the second pattern generation unit. A plurality of characters from which the first feature extraction unit extracts feature data from each of a plurality of character patterns of the first character pattern group generated by performing the first image processing. A step of extracting each feature data from a pattern, and a step in which the second feature extraction unit extracts feature data from each of a plurality of character patterns of a second character pattern group generated by performing the second image processing. The first similarity calculator calculating the similarity between the plurality of feature data extracted by the first feature extractor and the character feature data of the recognition dictionary; and the second feature extractor A step of calculating the similarity between the plurality of feature data extracted by the unit and the feature data of the characters in the recognition dictionary by the second similarity calculation unit; And a similarity calculated with the calculated degree of similarity by the second similarity degree calculating section by serial first similarity calculation unit integrated into one by the similarity integration section calculation formula predetermined integrated Selecting a character with a high similarity from the recognition dictionary using the similarity .

  ADVANTAGE OF THE INVENTION According to this invention, the precision which recognizes a character from a character image can be improved.

It is a figure which shows the structure of the image processing apparatus of one Embodiment of this invention. It is a flowchart which shows operation | movement of an image processing apparatus. It is a figure for demonstrating 4 neighborhood Gaussian filter. It is a figure for demonstrating an 8-neighbor Gaussian filter.

  Hereinafter, an image processing apparatus according to an embodiment of the present invention will be described in detail with reference to the drawings.

  As shown in FIG. 1, the image processing apparatus according to this embodiment includes an input unit 1, a computer 2 (hereinafter referred to as “PC2”), an output unit 3, and the like.

  The input unit 1 is an external input device such as a camera / scanner, and inputs a document image optically read from a paper document (document) by a CCD sensor or the like to the PC 2. The output unit 3 is, for example, a display device such as a monitor or a printing device such as a printer, and outputs (displays or prints) recognition result data output from the PC 2.

  The PC 2 includes a memory 10, a preprocessing unit 11, pattern generation units 12a and 12b, feature extraction units 13a and 13b, recognition dictionaries 14a and 14b, partial similarity calculation units 15a and 15b, a similarity integration unit 16, and the like. Yes. Each of these units is realized as a software module installed on a hard disk of a computer. Note that these units may be configured by hardware.

  The pattern generation unit 12a, the feature extraction unit 13a, the recognition dictionary 14a, and the partial similarity calculation unit 15a are the first calculation system 4. The first calculation system 4 changes the character image 22 generated by the preprocessing unit 11 into several character patterns 23 and 24, and then calculates the degree of similarity with the character 27a stored in the recognition dictionary 14a. To do.

The pattern generation unit 12b, the feature extraction unit 13b, the recognition dictionary 14b, and the partial similarity calculation unit 15b are the second calculation system 5. The second calculation system 5 changes the character image 22 generated by the preprocessing unit 11 into several character patterns 25 and 26 by a process different from that of the first calculation system 4, and then stores the character image 22 in the recognition dictionary 14a. The degree of similarity with the current character 27b is calculated.
The first calculation system 4 and the second calculation system 5 change the character image generated by the preprocessing unit 11 into several patterns by different processes for each system, and then change the character images to the corresponding recognition dictionaries 14a and 14b. It is a plurality of calculation systems for calculating the similarity with stored characters.

  The memory 10 is used not only as an area for reading a computer control program such as an operating system (OS) but also as a storage area for calculation and processing by the above-described units. The memory 10 stores, for example, image data for comparison processing, processing result data, and the like.

  The preprocessing unit 11 performs predetermined preprocessing such as partial segmentation, binarization, noise removal, and contour enhancement of an image (character image) used for character recognition. The pre-processing unit 11 performs at least one of character image partial segmentation, binarization, noise removal, contour enhancement, and the like as predetermined pre-processing. Since these individual image processing techniques are known techniques, a detailed description thereof will be omitted.

The pattern generation units 12a and 12b change the original character image by performing image processing processing such as expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation on the image preprocessed by the preprocessing unit 11. A new image that has been deformed or altered) is generated.
The pattern generation unit 12a generates a first character pattern group having a plurality of different character patterns 23 and 24 by performing a first image processing process on the character image. The pattern generation unit 12a performs, for example, a character (black pixel) movement process as the first image processing process.
The pattern generation unit 12b generates a second character pattern group having a plurality of different character patterns 25 and 26 by performing a second image processing process on the character image. The pattern generation unit 12b performs, for example, a character (black pixel) moving process and a character (black pixel) expansion process as the second image processing process. The character moving process is a process of shifting the position of a character (translating black pixels in parallel) within a possible range (character frame) of the character. The character expansion process is a process for thickening a character line in units of pixels.

  The feature extraction unit 13a extracts feature amounts (hereinafter referred to as feature data) of the individual character patterns 23 and 24 in the first character pattern group generated by the pattern generation unit 12a. The feature extraction unit 13b extracts feature amounts (hereinafter referred to as feature data) of the individual character patterns 25 and 26 in the second character pattern group generated by the pattern generation unit 12b.

  It is assumed that the character images 23 and 24 and the character images 25 and 26 are images obtained by performing different types of image processing. Different image processing processes may include the same processing process in part.

  In the recognition dictionaries 14a and 14b, a plurality of (many) characters and their feature data are stored in advance. The recognition dictionary 14a stores feature patterns (feature data and text data) of characters 27a for recognizing the character patterns 23 and 24 generated by the pattern generation unit 12a. The recognition dictionary 14b stores feature patterns (feature data and text data) of the characters 27b for recognizing the character patterns 25 and 26 generated by the pattern generation unit 12b.

The partial similarity calculation unit 15a obtains the similarity between the plurality of feature data extracted by the feature extraction unit 13a and the character feature data stored in the recognition dictionary 14a by calculation.
The partial similarity calculation unit 15b obtains the similarity between the plurality of feature data extracted by the feature extraction unit 13b and the character feature data stored in the recognition dictionary 14b by calculation. The calculation means that calculation is performed by adding feature data to the calculation formulas (functions) shown in the formulas (5) to (7) stored in the memory 10.

  The similarity integration unit 16 integrates the similarity calculated by the first calculation system 4 and the similarity calculated by the second calculation system 5 into one. More specifically, the similarity integration unit 16 integrates a plurality of partial similarities calculated by the partial similarity calculation units 15a and 15b, respectively. For the integration of the similarity, a similarity integration function σ represented by Expression (8) stored in the memory 10 is used.

  The operation of this image processing apparatus will be described below with reference to the flowchart of FIG. 2 and FIGS.

  When a document to be recognized is set in the input unit 1 such as a camera / scanner and a digital camera performs a shooting operation or a scanner performs a scanning operation, an image of the document is read by the input unit 1 to the PC 2. It is output as a digital image (referred to as “document image 21”).

  When the document image 21 output from the input unit 1 is input to the PC 2, the document image 21 is temporarily stored in the memory 10 by the preprocessing unit 11 (step S101 in FIG. 2).

  After storing the document image 21 in the memory 10, the preprocessing unit 11 reads the document image 21 from the memory 10, performs predetermined preprocessing on the read document image 21, and character image 22 to be subjected to character recognition. Is generated (step S102) and stored in the memory 10. The predetermined preprocessing is a predetermined process among image processes such as partial image segmentation, binarization, noise removal, and contour enhancement. The character image 22 generated by the predetermined preprocessing is a character such as “A”, for example.

  The pattern generation unit 12a reads the character image 22 from the memory 10, and performs a first image processing process on the read character image 22 (pre-processed image), thereby a plurality of different character pattern groups (FIG. 1). Character patterns 23 and 24) are generated (step S103) and stored in the memory 10. The first image processing process is a predetermined process among processes such as image expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation. A character pattern group generated by the first image processing is referred to as a first character pattern group.

The character pattern 23 is the character “A” moved to the upper left in the character frame with respect to the character image 22 (preprocessed image). The character pattern 24 is the character “A” that is moved to the lower left in the character frame with respect to the character image 22 (preprocessed image).
The pattern generation unit 12b reads the character image 22 from the memory 10 almost simultaneously with the pattern generation unit 12a, and performs a second image processing process on the read character image 22 (pre-processed image). A group of different character patterns (character patterns 25 and 26 in FIG. 1) is generated (step S104) and stored in the memory 10. A character pattern group generated by the second image processing is referred to as a second character pattern group.

The second image processing is a predetermined process among processes such as image expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation, and is different from the first image processing.
The character pattern 25 is a character “A” that is moved to the upper right in the character frame and is bolded with respect to the character image 22 (preprocessed image). The character pattern 26 is a character “A” that is moved to the lower right in the character frame and is bold with respect to the character image 22 (preprocessed image).

  That is, the first image processing and the second image processing are different predetermined processes among processes such as image expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation.

  The feature extraction unit 13a reads a first character pattern group, that is, a plurality of character patterns 23 and 24 from the memory 10, extracts respective feature data from the read character patterns 23 and 24 (step S105), and stores them in the memory 10 To do. At approximately the same time, the feature extraction unit 13b reads the second character pattern group, that is, the plurality of character patterns 25 and 26 from the memory 10, and extracts the respective feature data from the read character patterns 25 and 26 (step S106). Store in the memory 10.

  The partial similarity calculation units 15a and 15b read the feature data of the character patterns 23 and 24 from the memory 10, and use the plurality of read feature data and the feature data of the character 27a read from the recognition dictionary 14a to obtain the partial similarity. Is calculated (step S107) and stored in the memory 10.

  At substantially the same time, the partial similarity calculation unit 15b reads the feature data of each of the character patterns 25 and 26 from the memory 10, and uses the plurality of read feature data and the feature data of the character 27b read from the recognition dictionary 14b. The similarity is calculated (step S108) and stored in the memory 10.

  The similarity integration unit 16 reads the calculated partial similarity of each character pattern group from the memory 10 and integrates them (step S109).

  Then, using the integrated similarity, the similarity integration unit 16 selects characters with high similarity from the recognition dictionaries 14a and 14b (step S110), that is, performs a pattern recognition process, and recognizes characters (text data and text data) as a recognition result. The recognition source character image 22) is output to the output unit 3, and if the output unit 3 is, for example, a display device, the recognition result is displayed on the screen of the display device.

  Here, the character pattern generation processing (image processing) performed by the pattern generation units 12a and 12b will be described.

  The pattern generation units 12a and 12b perform predetermined image processing such as expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation of the image that has been preprocessed by the preprocessing unit 11, and the original character image 22 A plurality of new images are generated by changing (deformation or alteration) within the cutout range (character frame range), and the generated plurality of images are grouped (grouped) according to the image processing method.

  For example, the first group is obtained by performing an expansion process on a preprocessed image, the second group is obtained by performing a contraction process on a preprocessed image, and the third group is a rotation process performed on a preprocessed image. It is possible to divide into groups such as

  As an example of the image expansion processing, for example, for each pixel, if one or more of the pixels or four pixels in the upper, lower, left, and right directions is black, the processing is performed such that the pixels are also black.

  As an example of the image contraction process, for example, for each pixel, if one or more of the pixels or four pixels above, below, left, and right are white, the pixel is also white (contracted).

で表されるＲｕ［ｆ］，Ｓｓ［ｆ］をそれぞれ、回転行列Ｕおよび移動量ｓをパラメータとする回転済み、または移動済み画像とすることができる。この処理によって、座標Ｕ^-1ｘ,（ｘ−ｓ）にあった黒点が座標ｘの位置にそれぞれ回転・平行移動する。 Regarding rotation and movement of the image, when the pixel value of the coordinate x is represented by f (x) in the preprocessed image,

Ru [f] and Ss [f] represented by the above can be rotated or moved images using the rotation matrix U and the moving amount s as parameters, respectively. By this process, the black point at the coordinate U ⁻¹ x, (x−s) is rotated and translated to the position of the coordinate x.

  As for the rotation of the image, for example, it is rotated in increments of 10 degrees up to 90 degrees. As for the movement of the image, for example, the image is moved using a ratio such as 1/4 based on the length of the side or diagonal line of the preprocessed image.

  For image blurring and camera shake processing, a point spread function (PSF) for realizing these is prepared, convolution processing is performed on the preprocessed image, and binarization is performed again.

  As shown in FIG. 3, the point spread function (PSF) corresponding to the blur is obtained when the center pixel is set to “2” in the 9-pixel model in which the center pixel and the neighboring pixels in the eight directions are arranged. There is a 4-neighbor Gaussian filter in which the upper, lower, left and right pixels are set to “1” and the diagonal pixel is set to “0”.

  In addition, as shown in FIG. 4, in the 9-pixel model in which the center pixel and pixels adjacent to the surrounding 8 directions are arranged, when the center pixel is “4”, the upper, lower, left and right pixels are An 8-neighbor Gaussian filter or the like having “2” and a diagonal pixel “1” is used.

として作ることができる。点拡がり関数（ＰＳＦ）として１（ｙ）を原画像ｆ（ｘ）に畳み込む処理は、

と表せる。上記ｈ（ｘ）が畳み込み処理後の画像である。 As a point spread function (PSF) corresponding to camera shake, a point P is selected in the vicinity of the origin O,

Can be made as The process of convolving 1 (y) into the original image f (x) as a point spread function (PSF) is as follows:

It can be expressed. The h (x) is an image after the convolution process.

  Perspective transformation is widely known as the name of projective transformation. Projective transformation technology disclosed in general projective geometry literature such as Yumuro Kawamata "Geometry of Projective Space" (2001), Asakura Shoten, etc. Shall be used.

Processing for extracting feature values from pre-processed images (character patterns 23, 24 or 25, 26) by the feature extraction units 13a and 13b will be described below.
For example, there is a method in which a blurring process is performed on an image by the above-described method, and the blurred image is regarded as a vector having a pixel value as a component and used as a feature amount as it is. At this time, different feature extraction may be performed for each group of the images.

  In addition, as a method of performing pattern recognition processing by the partial similarity calculators 15a and 15b and a method of creating the recognition dictionaries 14a and 14b, the recognition dictionaries 14a and 14b are created based on the CLAFIC method disclosed in Non-Patent Document 2. Then, the similarity between the feature quantities of the plurality of generated patterns and the character types registered in the recognition dictionaries 14a and 14b is calculated using a mutual subspace method or the like.

を計算し、その固有ベクトルをｕ^α _１，ｕ^α _ｍとした上で、０≦ｐ≦ｍ，０≦ｑ≦ｎとなる整数ｐ、ｑを選んだ上で、行列Ｕ_ｐ＝（ｕ^α _１，…ｕ^α _ｐ），Ｖ_ｑ＝（ｖ_１…ｖ_ｑ）を用いて定義される

の最大固有値ρ^αを求め、このρ^αを類似度とする。ただし、左肩のtは転置を表す。このときρ^αは、例えば二宮市三編著「数値計算のわざ」（2006）、共立出版）などに開示されている累乗法などの既知の計算方法を用いて計算する。ただし、ｖ₁ ，ｖ_n は辞書データであり、これは各文字種毎に予め準備した学習パターンｙ¹，…，ｙⁿを用いて行列

を計算し、その固有ベクトルをｖ₁ ，ｖ_n とすることで計算できる。 As a similarity calculation method using the mutual subspace method, for example, an element of a group α of a plurality of feature vectors input from the feature extraction units 13a and 13b is used.

And the eigenvectors are set as u ^α ₁ and u ^α _m, and integers p and q satisfying 0 ≦ p ≦ m and 0 ≦ q ≦ n are selected, and then the matrix U _p = (u ^α ₁ ,... U ^α _p ), V _q = (v ₁ ... V _q )

The maximum eigenvalue ρ ^α is obtained, and this ρ ^α is used as the similarity. However, t on the left shoulder represents transposition. At this time, ^ρ α is, for example, Ninomiya City three written and edited by "numerical work of calculation" (2006), is calculated using the known calculation methods, such as exponentiation, which is disclosed in, for example, Kyoritsu Publishing). However, v ₁ and v _n are dictionary data, which is a matrix using learning patterns y ¹ ,..., Y ⁿ prepared in advance for each character type.

And the eigenvectors are set as v ₁ and v _n .

  In this example, the recognition dictionaries 14a and 14b prepare separate learning patterns for the partial similarity calculation units 15a and 15b corresponding to the pattern generation units 12a and 12b, respectively. A common recognition dictionary may be used in the degree calculation units 15a and 15b.

と定まるρを類似度とする。 As processing of the similarity calculation integration unit 16, when the partial similarity of each group is expressed as ρ ¹ ,..., Ρ ^μ , a certain similarity integration function σ is used.

Is defined as ρ.

  At this time, as a method of determining the similarity integration function σ, a method of selecting the largest of the partial similarities, a method of selecting some of the partial similarities in descending order and averaging them, and taking an average of the entire partial similarities There is a way. Further, instead of directly performing the averaging operation on the selected partial similarity, the difference in partial similarity may be emphasized once using a monotonically increasing function.

などとする方法がある。ただし、ｒは１≦ｒ≦μとなる整数であり、ρ^t(s)は、ρ¹，…，ρ^μのうち大きいものからs番目の値である。 To do so, using a monotonically increasing function τ

There is a method to say. However, r is an integer comprised between ^{1 ≦ r ≦ μ, ρ t} (s) is, ρ ^1, ..., a s th value from the largest of the [rho ^mu.

などが挙げられる。ただし、ρ₀ ，βは定数として適当なものを別途選ぶ。 Furthermore, as an example of τ

Etc. However, ρ ₀ and β are appropriately selected as constants.

  As another method of realizing the similarity integration unit 16, an existing method such as a neural network disclosed in Non-Patent Document 2 may be used. Further, in the partial similarity calculation units 15a and 15b and the similarity integration unit 16, when there is no character type with a significantly high similarity, it may be rejected as an unknown result.

  As described above, according to the image processing apparatus of this embodiment, the character (black pixel) of the character image to be recognized is positively changed partially (eg, the black pixel is shifted or thickened according to a predetermined rule). A plurality of character pattern groups including different character patterns generated are generated, and the partial similarity between the plurality of feature data of each character pattern group and the corresponding feature data of the recognition dictionaries 14a and 14b is calculated. Since the similarities are integrated into one, the character recognition accuracy can be improved.

In addition, this invention is not limited only to the said embodiment, You may deform | transform a component in the range which does not deviate from the summary in an implementation stage.
In the above-described embodiment, two calculation systems of the first calculation system 4 and the second calculation system 5 are illustrated. However, for example, a third calculation system and a fourth calculation system may be added. A plurality of calculation systems may be added, and a plurality of similarities obtained by calculation of each calculation system may be integrated.
In this case, the plurality of calculation systems are stored in the corresponding recognition dictionaries 14a and 14b after the character image generated by the preprocessing unit 11 is changed into several patterns by different processes for each system. The degree of similarity with the existing character is calculated. Further, the similarity integration unit 6 integrates a plurality of similarities obtained as calculation results by a plurality of calculation systems into one.

In addition, each component of the above embodiment may be realized by a program installed in a storage such as a hard disk device of a computer.
Furthermore, the program may be stored in a storage medium such as a computer-readable CD-ROM and the program may be read from the storage medium by a computer. Further, the configuration may be realized by distributing and storing components in different computers connected via a network, and communicating between computers in which the components are functioning.

  DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Computer (PC), 3 ... Output part, 4 ... 1st calculation system, 5 ... 2nd calculation system, 10 ... Memory, 11 ... Pre-processing part, 12a, 12b ... Pattern generation part, 13a , 13b ... feature extraction unit, 14a, 14b ... recognition dictionary, 15a, 15b ... partial similarity calculation unit, 16 ... similarity integration unit.

Claims (5)

A memory storing document images;
A recognition dictionary in which characters and their feature data are stored correspondingly;
A preprocessing unit that generates a character image by performing predetermined preprocessing on the document image read from the memory;
A first pattern generation unit that generates a first character pattern group having a plurality of different character patterns by performing a first image processing process on the character image generated by the pre-processing unit;
A second pattern generation unit that generates a second character pattern group having a plurality of different character patterns by performing a second image processing process on the character image generated by the pre-processing unit;
A first feature extraction unit that extracts feature data from each of a plurality of character patterns of the first character pattern group generated by the first pattern generation unit;
A second feature extraction unit that extracts feature data from each of a plurality of character patterns of the second character pattern group generated by the second pattern generation unit;
A first similarity calculator for calculating a similarity between a plurality of feature data extracted by the first feature extractor and character feature data stored in the recognition dictionary;
A second similarity calculation unit for calculating a similarity between a plurality of feature data extracted by the second feature extraction unit and character feature data stored in the recognition dictionary;
The similarity calculated by the first similarity calculator and the similarity calculated by the second similarity calculator are integrated into one by a predetermined calculation formula , and the integrated similarity is used. An image processing apparatus comprising: a similarity integration unit that selects characters with high similarity from the recognition dictionary . The image processing apparatus according to claim 1 .
The pre-processing unit is
An image processing apparatus that performs at least one of partial cutout, binarization, noise removal, and contour enhancement of the character image as predetermined preprocessing. The image processing apparatus according to claim 1 .
The first pattern generator is
As the first image processing, at least one of expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation is performed,
The second pattern generator is
An image processing apparatus that performs an image processing process different from the first image processing process among expansion, contraction, rotation, movement, blurring, camera shake, and perspective transformation as the second image processing process. The image processing apparatus according to claim 1 .
The similarity integration unit
The maximum similarity is selected from a plurality of similarities respectively calculated by the first similarity calculation unit and the second similarity calculation unit, some are selected in descending order of the similarity, and the similarities are averaged. An image processing apparatus that integrates the similarities by using any one calculation formula of taking an average of the whole and emphasizing a difference in the similarity using a monotonically increasing function . Memory in which document image is stored, recognition dictionary in which characters and their feature data are stored correspondingly, preprocessing unit, first pattern generation unit, second pattern generation unit, first feature extraction unit, second feature extraction unit In the image processing method by the image processing apparatus having the first similarity calculation unit, the second similarity calculation unit, and the similarity integration unit,
A step of reading a document image from the memory and performing a predetermined preprocessing on the read document image to generate a character image;
Generating a first character pattern group having a plurality of different character patterns by performing a first image processing on the character image by the first pattern generation unit;
Generating a second character pattern group having a plurality of different character patterns by performing a second image processing process on the character image by the second pattern generation unit;
The first feature extraction unit extracts feature data from each of the plurality of character patterns extracted from the plurality of character patterns of the first character pattern group generated by performing the first image processing. Steps,
The second feature extraction unit extracting feature data from each of a plurality of character patterns of the second character pattern group generated by performing the second image processing process;
A step in which the first similarity calculation unit calculates a similarity between the plurality of feature data extracted by the first feature extraction unit and character feature data of the recognition dictionary;
A step in which the second similarity calculation unit calculates the similarity between the plurality of feature data extracted by the second feature extraction unit and the character feature data of the recognition dictionary;
Integrated into one by the equation of the similarity integration section and a computed similarity by the the calculated similarity second similarity calculation unit by the first similarity calculation unit is predetermined, integrated And selecting a character with a high similarity from the recognition dictionary using the similarity .

Images