JP2010098744A - Image processing apparatus and image processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus which can compress an image according to information. <P>SOLUTION: The image processing apparatus is provided with: an analysis means which analyzes an input image; and a compressing means which compresses a first image which belongs to a first group based on the analysis result of the analysis means by a first compressing parameter, and compresses a second image which belongs to a second group by a second compressing parameter which has a possibility to deteriorate image quality than the first compressing parameter. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image compression technique, and more particularly, to a technique for converting a bitmap into a dictionary.

  Conventionally, in order to efficiently compress a binary image, JBIG2 has been standardized as a method of converting an image bitmap into a dictionary and using an index (corresponding dictionary number and its arrangement) and a dictionary as compressed data. On the other hand, when applying OCR to compression, image processing and signal conversion / reversible / irreversible are switched according to the accuracy of the OCR result (Patent Document 1), or a bitmap is dictionaryd in units of OCR characters in dictionary compression. (Patent Document 2), or a technique for changing a file format (CSV, hyperlink) to be converted if there is a table or there is a specific character in OCR (Patent Document 3).

  However, in a method in which OCR is applied to compression, there is no concept of importance of information itself, so that important information may be lost from both the OCR and the compressed image.

  An object of the present invention is to provide an image processing apparatus and an image processing method capable of compression according to information.

  An image processing apparatus according to an embodiment of the present invention compresses a first image belonging to a first group at a first compression rate based on an analysis unit that analyzes an input image, and an analysis result of the analysis unit, Compression means for compressing a second image belonging to the second group with a second compression parameter that may cause image degradation from the first compression parameter. The compression parameter is a parameter that can set the compression method and the degree of deterioration.

  An image processing method according to an embodiment of the present invention analyzes an input image, compresses a first image belonging to the first group based on the analysis result with a first compression parameter, and converts the first image belonging to the second group. The second image is compressed with a second compression parameter that may cause image degradation from the first compression parameter.

  ADVANTAGE OF THE INVENTION According to this invention, the image processing apparatus and the image processing method which can be compressed according to information can be provided.

1 is a block diagram illustrating a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 1-1 of the present invention. 3 is a block diagram illustrating an example of a schematic configuration of a layout analysis unit 1002. FIG. FIG. 10 is a diagram illustrating an example of the operation of a layout analysis unit 1002. FIG. 3 is a diagram illustrating an example of a schematic configuration of an image component converting unit 1003. It is a figure which shows an example of component image calculation. 5 is a diagram illustrating an example of a schematic configuration of a character recognition unit 1004. FIG. 5 is a diagram illustrating an example of a schematic configuration of a character analysis unit 1005. FIG. 5 is a diagram illustrating an example of a schematic configuration of an image dictionary unit 1006. FIG. It is a figure which shows an example of the keyword TABLE1005-2. It is a figure which shows an example of parameter TABLE1006-3. It is a block diagram which shows schematic structure of the image processing apparatus (image compression apparatus) which concerns on Embodiment 1-2 of this invention. It is a figure which shows an example of schematic structure of the character analysis part 1005A. It is a figure which shows an example of TABLE data of keyword TABLE1005A-2. It is a block diagram which shows schematic structure of the image processing apparatus (image compression apparatus) which concerns on Embodiment 2-1 of this invention. 3 is a block diagram illustrating an example of a schematic configuration of a layout analysis unit 2002. FIG. It is a figure which shows an example of the area | region detection operation | movement except the operation | movement of the object attribute determination part 2002-4. It is a figure which shows an example of schematic structure of the object attribute determination part 2002-4. It is a figure which shows an example of object determination data. An example of an object attribute determination operation will be described. 2 is a diagram illustrating an example of a schematic configuration of an image component converting unit 2003. FIG. It is a figure which shows an example of component image calculation. 3 is a diagram illustrating an example of a schematic configuration of an image dictionary unit 2004. FIG. It is a figure which shows the relationship between the object attribute information 2012 and the matching parameter 2004-5. It is a block diagram which shows schematic structure of the image processing apparatus (image compression apparatus) which concerns on Embodiment 2-2 of this invention. It is a figure which shows an example of schematic structure of the layout analysis part 2002A. It is a figure which shows an example of white background and non-white background determination. It is a figure which shows an example of schematic structure of the image dictionary formation part 2004A. It is a figure which shows the example which makes a non-white background quasi-reversible and makes a white background irreversible. It is a figure which shows that the character on a non-white ground tends to become unstable compared with the character on a white ground. It is a block diagram which shows schematic structure of the image processing apparatus (image compression apparatus) which concerns on Embodiment 2-3 of this invention. It is a figure which shows an example of schematic structure of the image dictionary formation part 2004B. It is a figure which shows an example of an input image. It is a figure which shows an example of the dictionary information corresponding to the input image shown in FIG. It is a figure which shows an example of an input image. It is a figure which shows an example of the image which compressed and restored this input image with the first character of the input image shown in FIG. It is a figure which shows an example of the image which compressed with the optimal character in the input image shown in FIG. 33, and decompress | restored this input image. It is a figure which shows the concept of the compression process by the image processing apparatus which concerns on Embodiment 3-1 of this invention. It is a figure which shows an example of the detailed structure of the image processing apparatus which concerns on Embodiment 3-1 of this invention. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 3-1 of this invention. It is a figure which shows an example of the compression process which concerns on Embodiment 3-1 of this invention. It is a figure which shows an example of an input image. It is a figure which shows the state 1 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 2 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 3 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 4 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 5 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows an example of the image which compressed and restored this input image with the first character of the input image shown in FIG. It is a figure which shows an example of the image which compressed with the optimal character in the input image shown in FIG. 40, and decompress | restored this input image. It is a figure which shows an example of a detailed structure of the image processing apparatus which concerns on Embodiment 3-2 of this invention. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 3-2 of this invention. It is a figure which shows an example of the image quality improvement of Symbol. It is a figure which shows schematic structure of the image processing apparatus which concerns on Embodiment 4 of this invention. It is a figure which shows the image of the decoding process by the decoding means of the image processing apparatus which concerns on Embodiment 4 of this invention. It is a figure which shows an example of the detailed structure of the image processing apparatus which concerns on Embodiment 4-1 of this invention. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 4-1 of this invention. It is a figure which shows an example of the compression process which concerns on Embodiment 4-1 of this invention. It is a figure which shows the state 1 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 2 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 3 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 4 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 5 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows the state 6 of the dictionary buffer corresponding to the input image shown in FIG. It is a figure which shows an example of the image which compressed and decompress | restored the input image shown in FIG. It is a figure which shows an example of the detailed structure of the image processing apparatus which concerns on Embodiment 4-2 of this invention. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 4-2 of this invention. It is a figure which shows an example of the compression process which concerns on Embodiment 4-2 of this invention. It is a figure which shows an example of an input image. FIG. 67 is a diagram showing an example of a restored image obtained by compressing and restoring the input image shown in FIG. 66 under a loose matching condition. FIG. 67 is a diagram showing an example of a restored image obtained by compressing and restoring the input image shown in FIG. 66 under strict matching conditions. FIG. 67 is a diagram showing an example of a restored image obtained by compressing and restoring the input image shown in FIG. 66 under appropriate matching conditions. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 5-1 of this invention. It is a figure which shows an example of an input image. It is a figure which shows the processing result of the character of the left end of the input image shown in FIG. FIG. 72 is a diagram illustrating a processing result of the second character from the left of the input image illustrated in FIG. 71. FIG. 72 is a diagram illustrating a processing result of the third character from the left of the input image illustrated in FIG. 71. It is a figure which shows the processing result of the character of the right end of the input image shown in FIG. It is a figure which shows the detail of the matching parameter determination part 5001 of the image processing apparatus which concerns on Embodiment 5-1 of this invention. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 5-2 of this invention. It is a figure which shows the processing result of the character of the left end of the input image shown in FIG. FIG. 72 is a diagram illustrating a processing result of the second character from the left of the input image illustrated in FIG. 71. FIG. 72 is a diagram illustrating a processing result of the third character from the left of the input image illustrated in FIG. 71. It is a figure which shows the processing result of the character of the right end of the input image shown in FIG. It is a figure which shows an example of an image compression apparatus. It is a figure for demonstrating an example of the technique regarding the precision improvement and speeding-up of template matching. It is a figure which shows an example of the whole structure of the image processing apparatus which concerns on Embodiment 6 of this invention. It is a figure which shows an example of the image compression using an interpolation image. It is a figure which shows an example of the image compression using an interpolation image. It is a figure which shows an example of the image compression using an interpolation image.

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

  FIG. 1 is a block diagram showing a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 1-1 of the present invention. As shown in FIG. 1, the image processing of the image processing apparatus is controlled by the control unit 1019. The scanner 1001 outputs an input image signal 1010 corresponding to an input image (original image). The layout analysis unit 1002 analyzes a layout of an object included in the input image signal 1010 using a predetermined layout analysis technique, and outputs object arrangement information 1011. The image component conversion unit 1003 converts the image signal 1010 into a component image 1012 using a predetermined image component conversion technique and object arrangement information 1011.

  The character recognition unit 1004 outputs a character code 1013 corresponding to the component image 1012 using a predetermined character recognition technique, and the character analysis unit 1005 calculates character analysis information 1014 from the character code 1013.

  The image dictionary unit 1006 outputs an image dictionary 1015 and a dictionary index 1015 corresponding to the component image 1012 using a predetermined bitmap dictionary technology and character analysis information 1014. The image encoding unit 1007 outputs code data 1017 corresponding to the image dictionary 1015 and the dictionary index 1015. The image file unit 1008 generates a searchable compressed file 1018 corresponding to the character code 1013 and the code data 1017.

  FIG. 2 is a block diagram illustrating an example of a schematic configuration of the layout analysis unit 1002. The reduction processing unit 1002-1 reduces the input image signal 1010 at a predetermined reduction rate, and outputs a reduced image 1002-4. The connected pixel search unit 1002-2 searches in eight directions using a predetermined chain algorithm, and outputs area information 1002-5 that is the upper left coordinates, lower left coordinates, upper right coordinates, and lower right coordinates of the areas where the pixels are connected. Since the coordinates of the area information 1002-5 are the coordinate system reduced by the reduction processing unit 1002-1, the area coordinate conversion unit 1002-3 converts the area information 1002-5 into the same coordinate system as the input image 1010. And output as object arrangement information 1011.

  FIG. 3 is a diagram illustrating an example of the operation of the layout analysis unit 1002. As shown in FIG. 3, it can be seen that the character area of the input image corresponding to the input image signal 1010 is generated as a part image in one block.

  FIG. 4 is a diagram illustrating an example of a schematic configuration of the image component converting unit 1003. The vertical pixel count unit 1003-1 and the horizontal pixel count unit 1003-2 divide the image signal 1010 in the coordinate unit of the object arrangement information 1011 and calculate the projection of the pixel count on the vertical axis and the horizontal axis, respectively. Are output as a vertical projection 1003-6 and a horizontal projection 1003-7. The comparator 1003-3 outputs a control signal 1003-8 to operate the selector 1003-4 so as to select the one with the larger variance of the projection values. The pixel dividing unit 1003-5 divides the input image signal 1010 using the selected projection value and outputs a component image 1012.

  FIG. 5 is a diagram illustrating an example of component image calculation. The image component converting unit 1003 compares the vertical projection 1003-6 and the horizontal projection 1003-7, selects a horizontal projection having a large variance, and the pixel dividing unit 1003-5 selects a threshold for the projection. By performing processing, the horizontal division coordinates are calculated (dotted line), and are output as a component image 1012 in the units indicated by “A”, “B”, “C”, “D”, “E”... Shown in FIG. .

  FIG. 6 is a diagram illustrating an example of a schematic configuration of the character recognition unit 1004. The character matching unit 1004-1 performs scaling / binarization / feature amount calculation on the input component image 1012, compares the feature amount with the data of the character code dictionary 1004-2, and has the highest degree of matching. The dictionary character code is output as the character code 1013.

  FIG. 7 is a diagram illustrating an example of a schematic configuration of the character analysis unit 1005. The keyword matching unit 1005-1 buffers the character code 1013 for a predetermined number of characters, reads and matches the keyword 1005-3 having a predetermined number of characters from the keyword TABLE 1005-2, and matches the keyword 1005- registered in the keyword TABLE 1005-2. 3 is output as character analysis information 1014.

  FIG. 8 is a diagram illustrating an example of a schematic configuration of the image dictionary unit 1006. The bitmap matching unit 1006-1 includes a bitmap dictionary 1006-2 in which the bitmap 1006-4 held in the dictionary process is registered, and a parameter TABLE 1006-3 in which a matching parameter 1006-5 is stored. Receive data from. The matching unit 1006-1 determines whether the component image 1012 exists in the bitmap dictionary 1006-2 by combining a predetermined position shift and matching processing, and if it exists, the dictionary index (dictionary index and image Position information of the component image) 1016 is output. If there is no corresponding bitmap pattern in the dictionary, the matching unit 1006-1 registers the component image 1012 in the dictionary and assigns an index.

  During this matching operation, the matching parameter 1006-5 corresponding to the attribute is read from the parameter TABLE 1006-3 according to the character analysis information 1014, so that compression according to the character information level is performed. Since the keyword matching unit 1005-1 buffers the predetermined number of characters and analyzes the character information as a character string, the bitmap matching unit 1006-1 also buffers the component image 1012 and analyzes it. It is configured to process in conjunction with the results.

  In this way, when dictionary formation is completed for all objects in the page of the input document, dictionary information is output as an image dictionary 1015.

  Then, the image encoding unit 1007 compresses the image dictionary 1015 with a predetermined image compression technique (for example, run length), outputs it together with the dictionary index 1016, and outputs it as code data 1017. The image file unit 1008 outputs the character code 1013 and the code data. A searchable compressed file 1018 consisting of 1017 is generated.

  9A and 9B are diagrams illustrating an example of a relationship between character analysis information and a matching accuracy parameter. As shown in FIG. 9A, information belonging to the first group is registered in the keyword TABLE1005-2. In other words, the keyword TABLE1005-2 includes “o”, “l”, “0”, “1”, which are easy to mistake numbers and alphabets, and “XYZ. C0Ltd "is registered, and character analysis information 1 and 2 and unregistered data are 0, respectively.

As shown in FIG. 9B, in parameter TABLE 1006-3, 0: irreversible (normal matching accuracy parameter) according to character analysis information 1014
1: Quasi-reversible (a parameter that gives results close to perfect match with slightly higher accuracy than usual)
2: Reversible (a parameter that considers only a perfect match)
It has become. For this reason, important information such as the company name is reversible (compression of the first compression parameter), and data that may be mistaken for OCR, such as numbers and alphabets, but is likely to be mistaken by the user is quasi-reversible (first The compression of the second compression parameter that may cause image degradation from the compression parameter of 1), and irreversible (compression of the third compression parameter that may cause image degradation from the second compression parameter) otherwise By changing the matching accuracy according to the character information level, it is possible to cover the deterioration of information due to compression and realize high compression. In this example, the company name is also registered twice in consideration of the possibility of mistakes in OCR. The compression parameter is a parameter that can set the compression method and the degree of deterioration.

  As described above, the compression parameter can be controlled in accordance with the level of information calculated from the OCR result, so that a highly compressed file with a balanced image quality and information can be generated.

  In the above-described embodiment, the compression method is described as a unified dictionary method. However, the lossless compression method can be directly compressed without using dictionary matching to improve the compression speed. .

  In addition, the character analysis information is composed of simple rules such as presence / absence based on keywords, but the analysis information is composed of, for example, a number or a numeric string (length is indefinite) following a specific keyword such as “¥” or “$”. ), It is possible to control information not as a simple number but as a concept of money.

  In OCR, the number of characters may increase or decrease compared to actual characters (for example, “XYZCoLtd”, etc.). However, it is possible to increase the certainty by registering keywords taking these into account. Using the uniqueness information, it is also possible to control the matching range for such candidates and increase the speed.

  Furthermore, according to the language type such as alphabetic characters such as alphabet, hiragana, kanji, etc., the configuration such as whether to include in the dictionary or not, whether the matching accuracy is changed, and the language type, for example, “m” or “ Control in consideration of the balance between image quality and information can be realized by controlling the balance between the frequency of occurrence of n ″ and the ease of error.

  In addition, although the data converted into parts by the image componentization unit is commonly used for character recognition and image dictionary conversion, if both target characters can be identified, it is naturally possible to take different configurations, such as layout analysis methods, A recognition method, dictionary formation, compression method, and the like are not limited to the present embodiment, and application methods such as a parameter control method and a reversible lossy method are not limited to the present embodiment.

  FIG. 10 is a block diagram showing a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 1-2 of the present invention. In the image processing apparatus shown in FIG. 10, the same reference numerals as those assigned to the image processing apparatus in FIG.

  The character analysis unit 1005A, the character analysis information 1014A output from the character analysis unit 1005A, the image file unit 1008A, and the searchable compressed file 1018A are slightly different from the embodiment 1-1. Further, an MMR compression unit 1009A and full-surface compression data 1019A are added.

  FIG. 11 is a diagram illustrating an example of a schematic configuration of the character analysis unit 1005A. The keyword TABLE1005A-2 is different from the embodiment 1-1. FIG. 12 is a diagram illustrating an example of TABLE data of the keyword TABLE1005A-2. As shown in FIG. 12, the character analysis information 1014A is shown in four levels of 0-3. In the embodiment 1-1, there are 3 levels from 0 to 2, but in the embodiment 1-1, there are 4 levels from 0 to 3. In the keyword TABLE 1005A-2, “Confidential” and a keyword with a high risk of being mistaken in the OCR of the word “Confidential” are registered, and level 3 (character analysis information 1014A) is set for these registered keywords.

  The MMR compression 1009A reversibly compresses the entire image as a full binary image by a predetermined binary compression method, and outputs it as full compressed data 1019A. The image file unit 1008A generates a searchable compressed file 1018A from the code data 1017 and the character code 1013, as in the case of Embodiment 1-1, except when the character analysis information 1014A indicates “3”. When the character analysis information 1014A indicates “3”, a searchable compressed file 1018A is generated from the entire compressed data 1019A and the character code 1013.

  As described above, a compressed file can be generated with a compression method that does not have a risk of information degradation, instead of a compression method that has a risk of information degradation, such as “Confidential”. Compressed files that take into account can be provided.

  The above-described Embodiments 1-1 and 1-2 are summarized below.

  (1) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and image compression that can control compression performance such as reversible and irreversible using the image as a parameter. The image processing apparatus includes a character analysis unit that analyzes the character information, and the image compression unit controls compression performance according to the character analysis result. Since the compression rate is controlled by the character recognition result, high image quality and high compression can be realized.

  (2) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, an image componentization unit that converts the image into a component and outputs a component image, and the component image An image dictionary, an image dictionary that outputs an image dictionary and a dictionary index of the component image, and an image encoding unit that encodes the image dictionary and the dictionary index, A character analysis unit that analyzes character information is included, and the image dictionary unit controls dictionary accuracy according to the character analysis result. Since the matching level of the dictionary system is controlled by the character recognition result, high image quality and high compression can be realized.

  (3) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, an image componentization unit that converts the image into a component and outputs a component image, and the component image An image dictionary and an image dictionary that outputs an image dictionary and a dictionary index of the component image or outputs it as an independent image without dictionary, and an image encoding that encodes the image dictionary, the dictionary index, and the independent image The image processing apparatus has a character analysis unit that analyzes the character information, and the image dictionary unit selectively outputs an image dictionary or an independent image according to the character analysis result. Since index / non-index encoding is switched according to the character recognition result, high image quality, high compression, and high speed can be realized.

  (4) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and image compression that can control compression performance such as reversible and irreversible using the image as a parameter. An image processing apparatus comprising: a character analysis unit that analyzes the character information, wherein the image compression unit controls compression performance according to the character analysis result; and Is a specific character such as a keyword or character type. Since the probability that a specific keyword is reversible in the character recognition result is increased, high image quality and high compression can be realized.

  (5) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and image compression that can control compression performance such as reversible and irreversible using the image as a parameter. An image processing apparatus comprising: a character analysis unit that analyzes the character information, wherein the image compression unit controls compression performance according to the character analysis result; and Is a specific character such as a keyword or character type and a character similar to the character. Since the compression is controlled by a specific keyword and data approximated by the character recognition result and other data, high image quality and high compression can be realized in consideration of OCR accuracy.

  (6) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and image compression that can control compression performance such as reversible and irreversible using the image as a parameter. An image processing apparatus comprising: a character analysis unit that analyzes the character information, wherein the image compression unit controls compression performance according to the character analysis result; and Is a language type. Since compression is controlled according to the language type in the character recognition result, control according to the characteristics of characters expressing languages such as English and Hiragana can be performed, so that high image quality and high compression can be realized.

  (7) An image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and image compression that can control compression performance such as reversible and irreversible using the image as a parameter. An image processing apparatus comprising: a character analysis unit that analyzes the character information, wherein the image compression unit controls compression performance according to the character analysis result; and Is the language type and the appearance frequency of the character in the language. Since the compression is controlled according to the language type in the character recognition result, it is possible to control according to the character expressing the language such as English and hiragana and the character of the language, so that high image quality, high compression and high speed can be realized.

  (8) The image processing apparatus includes an image input unit that inputs an image, a character recognition unit that recognizes the image and outputs character information, and an image compression unit that can compress the image using different compression methods. The image processing apparatus includes a character analysis unit that analyzes the character information, and the image compression unit controls a compression method according to the character analysis result. Since the matching level of the dictionary system is controlled by the character recognition result, high image quality and high compression can be realized.

  As described above, for example, a bitmap to which a character code extracted such as “l” or “1” that is likely to be lost in both the specific keyword and OCR and dictionary matching from the character code information of the character recognition result is losslessly compressed. Furthermore, dictionary accuracy is controlled in accordance with the type of language, and kanji characters with a large number of characters are removed from the dictionary target and highly compressed.

  As described above, since the image quality can be controlled according to the extracted character information, it is possible to avoid missing important information. Important information includes registered keywords, numbers, and letters and numbers that are easily mistaken. Also, to improve the compression rate, dictionary matching is not performed with kanji, kanji itself, or languages with a large number of strokes. Alternatively, the language is determined, and the accuracy of characters frequently appearing in the language is loosened (high compression), strict (high image quality), or index replacement (high speed) is performed. In the lexicographic method, the matching accuracy is made strict (reversible lexicographic) or excluded from the lexicographic target (non-lexicographic compression). Further, when a registered keyword is found, not only JBIG2 but also the overall compression rate is lowered, or the entire image is processed with JPEG.

  Hereinafter, non-dictionary and lexicization will be exemplified.

爨 → Non-dictionary because of the large number of strokes (judged by character code) Circle → License-based because the number of strokes is small ○, o, O, 0 → If matching is wrong (both OCR and dictionary), meaning is different and non-dictionary is reversible Compression 1, 2, 3 ... → Mismatching leads to a serious mistake, so it is lexicographically and losslessly compressed. Characters before and after $, $ → Mismatching leads to a serious mistake.・ Company name / person name → Mismatching will lead to a serious mistake, so it will be lexicographically and losslessly compressed Alphabet / Number / Hiragana / Kana → Dictionary, Kanji → Nondictionary Note that the above lossless compression is the first compression This is compression of parameters, and is compression that can reproduce an original image in a state close to 100%. The quasi-reversible compression is compression of the second compression parameter that may cause image degradation from the first compression parameter, and is compression that can reproduce the original image almost faithfully. The lossy compression is compression of a third compression parameter that may cause image degradation from the second compression parameter, and is compression that can reproduce the original image approximately faithfully. The compression parameter is a parameter that can set the compression method and the degree of deterioration.

  FIG. 13 is a block diagram illustrating a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 2-1 of the present invention. As shown in FIG. 13, the image processing of the image processing apparatus is controlled by the control unit 2006. The scanner 2001 outputs an input image signal 2010 corresponding to an input image (original image). The layout analysis unit 2002 outputs object arrangement information 2011 and object attribute information 2012 corresponding to the image signal 2010 input from the scanner 2001 using a predetermined layout analysis technique.

  The image component converting unit 2003 uses the object arrangement information 2011 to output a component image 2013 corresponding to the input image signal 2010.

  The image dictionary unit 2004 outputs an image dictionary 2014 and a dictionary index 2015 corresponding to the component image 2013 using a predetermined bitmap dictionary forming technique and object attribute information 2012, and the image encoding unit 2005 outputs the code data 2016. Generate and output.

  FIG. 14 is a block diagram illustrating an example of a schematic configuration of the layout analysis unit 2002. The reduction processing unit 2002-1 reduces the input image signal 2010 at a predetermined reduction rate, and outputs a reduced image signal 2002-6. The connected pixel search unit 2002-2 searches in eight directions using a predetermined chain algorithm, and outputs area information 2002-7 that is the upper left coordinates, lower left coordinates, upper right coordinates, and lower right coordinates of the areas where the pixels are connected. Since the coordinates of the region information 2002-7 are the coordinate system reduced by the reduction processing unit 2002-1, the region coordinate conversion unit 2002-3 converts the region information 2002-7 into the same coordinate system as the input image 2010. And output as object arrangement information 2011.

  On the other hand, the object attribute determination unit 2002-4 aggregates the area information 2002-7 in the page of the input image, compares it with the object determination data 2002-8 read from TABLE2002-5, and outputs object attribute information 2012. .

  FIG. 15 is a diagram illustrating an example of the region detection operation excluding the operation of the object attribute determination unit 2002-4.

  By reducing the input image 2010, it can be seen that in the reduced image 2002-6, pixels and the like are connected. When a connected region is calculated for this image using a predetermined chain algorithm, a plurality of regions 2002-7 for each cluster as shown by the dotted lines in FIG. 15 are calculated. If the coordinates of the region 2002-7 are converted, the object arrangement information 2011 in which the coordinate system is returned to before the reduction is calculated.

  FIG. 16 is a diagram illustrating an example of a schematic configuration of the object attribute determination unit 2002-4. Using the region information 2002-7, the height 2002-4-4 = H for each region is obtained as follows.

H = MIN (| upper left X coordinate−lower right X coordinate |, | upper left Y coordinate−lower right Y coordinate |)
The median height of each object in the page of the input image is calculated as the median region height 2002-4-5.

  The area distance calculation unit 2002-4-2 obtains the center coordinates 2002-4-6 of each area, obtains the nearest center coordinates 2002-4-7 and 2002-4-8 for each center coordinate, and determines the judgment part 2002. Output to -4--3.

  The determination unit 2002-4-3 takes the difference between the height 2002-4-4 and the median height of the region 2002-4-5 for each region, and calculates whether the difference is greater than the predetermined value as the region height difference.

  The center code and the center difference are calculated by the following equations, respectively.

Difference A = 2002-4-6−2002-4-7
Difference B = 2002-4-6−2002-4-8
Center code = difference A × sign of difference B center difference = | | difference A | − | difference B |
That is, the center code is negative when there are objects close to the top and bottom of the processing area, or right and left, and is positive when the objects are arranged in the same direction as two below.

  As the center difference, a state is calculated in which the difference in distance between each object and the target object is large or different.

  Therefore, the object attribute data shown in FIG. 17 is used to determine the attribute of the object and output as object attribute information 2012.

  FIG. 18 shows an example of an attribute determination operation for an object. Each diagonal line in FIG. 18 is an object to be determined, and obj1 and obj2 are the two closest objects selected. In FIG. 18 (a) and FIG. 18 (b), are the heights higher than other objects? The title is highly likely to belong to the highest position.

  In FIG. 18C, the height is small and the arrangement is within other objects, but the distance between the two objects is different, so there is a high possibility of a headword such as a paragraph change.

  Since FIG. 18D is neither, there is a high possibility of a normal text.

  FIG. 19 is a diagram illustrating an example of a schematic configuration of the image component converting unit 2003. The horizontal pixel count unit 2003-2 and the horizontal pixel count unit 2003-2 divide the image signal 2010 in the coordinate unit of the object arrangement information 2011, calculate the projection of the pixel count on the vertical axis and the horizontal axis, Output as a vertical projection 2003-6 and a horizontal projection 2003-7.

  The comparator 2003-3 outputs a control signal 2003-8 to operate the selector 2003-4 so as to select the one having the larger variance of the projection values.

  The pixel dividing unit 2003-5 divides the image 2010 using the selected projection value and outputs a component image 2013.

  FIG. 20 is a diagram illustrating an example of component image calculation. The image componentizing unit 2003 compares the vertical projection 2003-6 and the horizontal projection 2003-7, selects a horizontal projection with a large variance, and the pixel dividing unit 2003-5 sets a threshold for the projection. By performing processing, horizontal division coordinates are calculated (dotted line), and output as a component image 1012 in the units indicated by “A”, “B”, “C”, “D”, “E”,. .

  FIG. 21 is a diagram illustrating an example of a schematic configuration of the image dictionary unit 2004. The bitmap matching unit 2004-1 includes a bitmap dictionary 2004-2 in which the bitmap 2004-4 stored in the lexicization process is registered, and a parameter TABLE 2004-3 in which a matching parameter 2004-5 is stored. Receive data from. The bitmap matching unit 2004-1 determines whether the component image 2013 exists in the bitmap dictionary 2004-2 by combining a predetermined position shift and matching processing, and if it exists, the dictionary index (the index of the dictionary) (Position information of part image on image) 2015 is output. If there is no corresponding bitmap pattern in the dictionary, the bitmap matching unit 2004-1 registers the component image 2013 in the dictionary and assigns an index.

  At the time of this matching operation, the matching parameter 2004-5 corresponding to the attribute is read from the TABLE 2004-3 according to the object attribute information 2012, whereby the compression corresponding to the object attribute is performed. For example, as shown in FIG. 22, since it is highly likely that the information is a title or a headword, the parameter 2004-5 that makes the matching accuracy stricter is read, and otherwise, the normal matching accuracy parameter 2004- 5 is output.

  As described above, when dictionary formation for all objects in the page of the input document is completed, dictionary information is output as an image dictionary 2014.

  Then, the image encoding unit 2005 compresses the image dictionary 2014 with a predetermined image compression technique (for example, run length), and outputs it as code data 2016 together with the dictionary index 2015.

  As described above, for example, important information such as titles shown in FIGS. 18A, 18B, and 18C is compressed at a lossless or low compression rate, and data whose information is slightly reduced is subjected to higher compression processing. Therefore, it is possible to generate a highly compressed file that balances image quality and information deterioration.

  In the present embodiment, titles and headwords are given as examples. However, the attribute information is not limited to the present embodiment. For example, when a table is extracted, the inside may be more important data. Therefore, the compression rate may be changed similarly to other objects.

  Note that the layout analysis method, the object attribute information calculated therefrom, the dictionary formation, the compression method, and the like are not limited to the present embodiment, and the application method such as the parameter control method and the reversible irreversible is also the present embodiment. It is not limited to.

  FIG. 23 is a block diagram illustrating a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 2-2 of the present invention. In the image processing apparatus shown in FIG. 23, the same reference numerals as those assigned to the image processing apparatus in FIG. 13 are assigned to the blocks common to the embodiment 2-1.

  The layout analysis unit 2002A, the object density attribute 2012A, and the image dictionary unit 2004A are slightly different from the embodiment 2-1.

  FIG. 24 is a diagram illustrating an example of a schematic configuration of the layout analysis unit 2002A. The configuration of the layout analysis unit 2002A is basically the same as the configuration of the layout analysis unit 2002, except that it has an area density determination unit 2002-6A and outputs an object density attribute 2012A. The area density determination unit 2002-6A calculates a histogram in the area as shown in FIG.

  FIG. 26 is a diagram illustrating an example of a schematic configuration of the image dictionary unit 2004A. The image dictionary unit 2004A calculates the matching accuracy parameter 2004-5 using the object attribute information 2012 and the object density attribute 2012A.

  For example, as shown in FIG. 27, a non-white background is quasi-reversible and a white background is irreversible. That is, the compression rate is changed depending on whether the object is white. This is because, as shown in FIG. 28, the background of non-white ground characters is often expressed as a knitting point, so when binarized and a bitmap is generated, the shape is likely to be unstable compared to a white background. Deterioration is prevented by setting the compression rate of non-white background to a low level.

  If the emphasis is placed on efficient compression from the opposite viewpoint, the binarized bumps in FIG. 28 (b) are not information, and the matching accuracy is relaxed, and the information on the non-white background can be efficiently compressed. Is also obvious.

  In addition, it is possible to preferentially register a bit map of characters on the white background in the bitmap dictionary 2004-2, and it is also possible to make non-white characters easy to read.

  FIG. 29 is a block diagram showing a schematic configuration of an image processing apparatus (image compression apparatus) according to Embodiment 2-3 of the present invention. In the image processing apparatus shown in FIG. 29, the same reference numerals as those given to the image processing apparatus in FIG.

  An image bitmap 2017 that is an output of the image dictionary 2004B and the image dictionary 2004B is slightly different from the embodiment 2-1.

  FIG. 30 is a diagram illustrating an example of a schematic configuration of the image dictionary unit 2004B. The determination unit 2004-8B receives the object attribute information 2012. If the object attribute information 2012 is a title or a heading, the determination unit 2004-8B operates the selector 2004-9B to output the component image 2013 as an image bitmap 2017, and the other than the title and the heading. If it is (Others), it inputs into the matching part 2004-1, and a dictionary matching process is performed like Embodiment 2-1.

  The image encoding unit 2005B compresses the image dictionary 2014 and the image bitmap 2017 using the lossless compression technique in the same manner as the embodiment 2-1, and combines the position information of the dictionary index 2015 and the image bitmap 2017 to output as code data 2016B. To do.

  As described above, the object information to be reversibly processed is directly reversibly compressed without being converted into a dictionary, so that the compression speed can be increased.

  In the present embodiment, the image dictionary 2014 and the image bitmap 2017 are shown using the same lossless compression. However, both compression methods may be changed or the compression parameters may be changed.

  The above-described Embodiments 2-1, 2-2, and 2-3 are summarized below.

  (1) An image processing apparatus includes an image input unit that inputs an image, a layout analysis unit that outputs object arrangement information from the image, and an image component that converts the image into a component from the image and the object arrangement information and outputs a component image An image processing apparatus having an encoding unit and an image encoding unit that compresses the component image, wherein the layout analysis unit outputs an object attribute, and the image encoding unit corresponds to the object attribute Control the compression rate. Since the compression rate can be controlled based on the layout analysis result, it is possible to provide a compressed file with high image quality and high compression.

  (2) An image processing apparatus includes an image input unit that inputs an image, a layout analysis unit that outputs object arrangement information from the image, and an image component that converts the image into a component from the image and the object arrangement information and outputs a component image An image comprising: a conversion unit, an image dictionary converting the component image into an image dictionary and outputting a dictionary index of the component image, and an image encoding unit encoding the image dictionary and the dictionary index In the processing apparatus, the layout analysis unit outputs an object attribute, and the image dictionary unit controls dictionary accuracy according to the object attribute. Since dictionary accuracy can be controlled based on layout analysis results, compressed images with high image quality and high compression can be provided.

  (3) An image processing apparatus includes an image input unit that inputs an image, a layout analysis unit that outputs object arrangement information from the image, and an image component that converts the image into parts from the image and the object arrangement information and outputs a component image An image comprising: a conversion unit, an image dictionary converting the component image into an image dictionary and outputting a dictionary index of the component image, and an image encoding unit encoding the image dictionary and the dictionary index In the processing apparatus, the layout analysis unit outputs an object attribute, and the image dictionary unit controls dictionary accuracy according to the object attribute. The object attribute is a document heading, a table, or the like. Since dictionary accuracy can be controlled based on object attributes such as headings and tables, compressed files with high image quality and high compression can be provided.

  (4) An image processing apparatus includes: an image input unit that inputs an image; a layout analysis unit that outputs object arrangement information from the image; and an image component that converts the image into parts from the image and the object arrangement information and outputs a component image An image comprising: a conversion unit, an image dictionary converting the component image into an image dictionary and outputting a dictionary index of the component image, and an image encoding unit encoding the image dictionary and the dictionary index In the processing device, the layout analysis unit outputs an object attribute, and the image dictionary unit controls dictionary accuracy according to the object attribute, and the object attribute is the presence or absence of a background. Since the dictionary accuracy can be controlled according to the region attribute to which the character belongs, such as whether the character is on the background, a compressed file with high image quality and high compression can be provided.

  (5) An image processing apparatus includes an image input unit that inputs an image, a layout analysis unit that outputs object arrangement information from the image, and an image component that converts the image into parts from the image and the object arrangement information and outputs a component image An image comprising: a conversion unit, an image dictionary converting the component image into an image dictionary and outputting a dictionary index of the component image, and an image encoding unit encoding the image dictionary and the dictionary index In the processing apparatus, the layout analysis unit outputs an object attribute, and the image dictionary unit preferentially converts an object attribute including a white background into the image dictionary. Since characters on the background are preferentially used as a dictionary bitmap, a compressed file with high image quality and high compression can be provided.

  (6) An image processing apparatus includes: an image input unit that inputs an image; a layout analysis unit that outputs object arrangement information from the image; and an image component that converts the image into parts from the image and the object arrangement information and outputs a component image An image comprising: a conversion unit, an image dictionary converting the component image into an image dictionary and outputting a dictionary index of the component image, and an image encoding unit encoding the image dictionary and the dictionary index In the processing apparatus, the layout analysis unit outputs an object attribute, and the image dictionary unit controls whether to perform the dictionary compression according to the object attribute. Since the dictionary compression method is controlled based on the layout analysis result, a compressed file with high image quality and high compression can be provided.

  As described above, according to the result of the layout analysis processing for realizing JBIG2 and OCR, the area information and the document type are determined, and the matching accuracy of lexicographic compression and the application of non-lexicographic compression are controlled, so that high speed is achieved. A dictionary compression file with high image quality and high compression can be provided. In other words, important areas are determined according to the layout analysis results, and the compression rate is controlled, so a high-quality, high-compression compressed file can be generated at a high speed. Can be provided as a compressed file. That is, since the image quality can be controlled according to the extracted area information, it is possible to avoid missing important information. For example, the important information includes headings such as document headings, tables, tables, and graphs. In order to improve the overall accuracy, the quality of the character image generated by binarization or the like can be complemented by switching the compression parameter or the like depending on whether the character is on the ground. Alternatively, the base bitmap of the dictionary can be selected centering on white ground characters. For example, an area determined to be a title / headline is reversibly compressed so that the information can be reliably viewed as an image file, and OCR accuracy is ensured when the dictionary compression file is subjected to OCR. When a table is extracted, it is highly possible that the table area is reversibly compressed, or there is a high possibility of having only numbers, so the matching parameters are changed or the compression method is changed.

  Note that the above-described lossless compression is compression of the first compression parameter, and is compression that can reproduce the original image in a state close to 100%. The quasi-reversible compression is compression of the second compression parameter that may cause image degradation from the first compression parameter, and is compression that can reproduce the original image almost faithfully. The lossy compression is compression from a third compression parameter that may cause image degradation more than the second compression parameter, and is compression that can reproduce the original image approximately faithfully.

  Next, a third embodiment of the present invention will be described.

  There is a method of compressing an image by forming one dictionary bitmap for each character or the like and storing the dictionary bitmap and position information. However, the dictionary bitmap registered as a dictionary only registers a new bitmap that appears first, and does not necessarily select a bitmap having an optimal shape as a dictionary.

  Many compression methods are conceivable as techniques for compressing data. There are entropy coding and arithmetic coding represented by Huffman coding. As a pre-processing, there is a technique called universal coding (dictionary / dictionary-based coding) (hereinafter referred to as “dictionary”).

  LZ77 and LZ78, which are data compression algorithms developed by Jacob Ziv and Abraham Lempel, are compression methods that preserve the occurrence position and length of the existing characters. This is basically a character code compression method.

  JBIG2 SymbolDictionary, a binary compression technology that is an international standard in ISO / IEC14492, applies this mechanism to images. SymbolDictionary is a method in which an image area is regarded as one dictionary BMP (for example, one character BMP), common ones are handled as the same dictionary BMP, and compression is performed by having the dictionary BMP and position information. These are also effective for images having a specific pattern (such as character images and halftone images).

  A specific example of the dictionary compression of JBIG2 SymbolDictionary is shown below.

  For example, when there is an input image “ABCBAD” as shown in FIG. 31, “ABCBAD” is normally compressed as an entire image. In the dictionary compression, there are two images “A”. Therefore, only one image “A” is held, and the data is saved by having position information after that. In the case of such data compression, the following data is created by creating a dictionary.

Dictionary (Symbol) 4 types: “A” “B” “C” “D” (FIG. 32)
Dictionary (Symbol) position information 6 types: (image A: position (0, 0)), (image B: position (6, 0)), (image C: position (12, 0)), (image B: position (18,0)), (Image A: Position (24,0)), (Image D: Position (30,0))
In creating these data, lexicographic object image extraction processing (what is used as a dictionary candidate) is required (concatenated pixel extraction / character extraction). It is determined whether or not the extracted dictionary candidate can be determined to be the same as the existing dictionary (Symbol). If it is the same, the dictionary (Symbol) position information is registered, and if it is determined to be different, the dictionary (Symbol) information and the dictionary ( Symbolization processing having a mechanism for registering symbol position information is performed, and finally high compression is realized by compressing the dictionary information and the dictionary position information.

  In a dictionary compression method that considers the same dictionary (Symbol) with a pixel matching rate that is not 100%, an image that is initially registered as a dictionary (Symbol) is directly reflected in the output image. For this reason, if there is a problem with the image once registered as a dictionary (for example, a part of characters is missing), the defective dictionary (Symbol) is reflected in the output result.

  As a countermeasure, there is a method of having a font dictionary (Symbol) in advance, using the dictionary (Symbol) if there is a corresponding one, and vectorizing (quasi-reversible) if there is one (Japanese Patent Laid-Open No. 2005-208772) . However, the vectorization of characters raises concerns about a decrease in compression rate.

  Furthermore, an example of the above-described compression / decompression process will be described.

  FIG. 33 is a diagram illustrating an example of an input image. For example, a case is assumed where the image processing apparatus searches for dictionary candidate images from left to right and from top to bottom, and creates a dictionary bitmap. Dictionaries are irreversible and some pixel differences are considered the same dictionary. In this case, the image is compressed and restored as follows.

  (1) The first discovered character (left) is registered as a dictionary.

  (2) The center is determined to be the same character as the left character.

  (3) The right character is determined to be the same character as the left character.

  (4) As shown in FIG. 34, the restored image of the data created in this way has a shape in which three characters on the left of the input image are arranged.

  Therefore, in Embodiment 3 of the present invention, a certain amount of dictionary candidate images are held, and a dictionary (Symbol) that has a high degree of matching with other dictionary candidate images determined to be the same from the dictionary candidate images. By storing it as information, it is possible to reduce image quality defects such as missing characters. That is, in Embodiment 3 of the present invention, a bitmap for determining whether to register as a dictionary is stored, and a better bitmap is registered as a dictionary. As a result, an image closer to the input image can be maintained without depending on the lexicographic processing order (without depending on the bitmap shape initially registered in the dictionary).

  Specifically, the bitmap group determined to be the same dictionary is determined as follows to determine the dictionary BMP.

  (1) A pixel having the smallest change point between the white pixel and the black pixel is employed.

  (2) The one with the least noise pixels is adopted.

  (3) OCR is determined and the one with the highest degree of matching is adopted.

  As a result, an image closer to the input image can be maintained without depending on the lexicographic processing order (without depending on the bitmap shape initially registered in the dictionary).

For example, the image processing apparatus (image compression apparatus) according to the third embodiment of the present invention can compress and restore an image as follows. (1) A dictionary of the first character (left) found Register as

  (2) The center character is determined to be the same character as the left character. The dictionary candidate image (center character) is stored in a buffer.

  (3) The right character is determined to be the same character as the left character. The dictionary candidate image (right character) is stored in the buffer.

  (4) When the dictionary collection is completed, select the character with the least noise among the characters determined to be the same character (center character) and register it as a dictionary bitmap (high pixel connectivity: run is most connected) ing).

  (5) As shown in FIG. 35, the restored image has a shape in which three characters at the center of the input image are arranged.

  Hereinafter, Embodiment 3 of the present invention will be described in detail.

  FIG. 36 is a diagram showing a concept of compression processing by the image processing apparatus (image compression apparatus) according to Embodiment 3-1 of the present invention, and FIG. 38 is an image processing apparatus according to Embodiment 3-1 of the present invention. FIG. 37 is a diagram illustrating an example of an overall configuration of an (image compression apparatus), and FIG. 37 is a diagram illustrating an example of a detailed configuration of an image processing apparatus (image compression apparatus) according to Embodiment 3-1 of the present invention. FIG. 39 is a diagram illustrating an example of compression processing according to Embodiment 3-1 of the present invention. The image processing device shown in FIG. 36 corresponds to the image processing device shown in FIGS. 37 and 38, and mainly refers to the image processing device shown in FIGS. 37 and 38 and performs the compression processing according to Embodiment 3-1. explain.

  The Symbol match determination unit 3001 compares / determines whether the lexicographic object data D3002 and the symbol information I3004 existing in the dictionary buffer 3005 match, and determines a match determination result R3003 (if not match, It has a symbol comparison unit 3001-1 and a symbol comparison result output unit 3001-2 for outputting information (if matching, the ID number of the matching symbol).

  The Symbol match determination unit 3001 includes a means / method for determining a match by a known means / method, and that 100% pixels do not always have to be matched, and those that are somewhat similar are the same.

  The operation of the symbol / matching symbol information / symbol position information registration unit 3002 will be described. An ID is assigned to the lexical object data D3002, and the lexical object data D3002 and its ID are registered in the symbol information buffer 3005-1 of the dictionary buffer 3005. The dictionary buffer 3005 stores the ID and position information. Symbol position information buffer 3005-2 of ID / Symbol position information registration unit 3002-2 to register, and if the match determination result R3003 matches with other Symbol (Symbol candidate image) A matching symbol information registration unit 3002-3 for registering the ID number of the existing symbol in the matching symbol information buffer 3005-3 of the dictionary buffer 3005.

  The operation of the symbol determination unit 3003 will be described. When the symbol correction flag specified by the user is ON (when the user wants to optimize the symbol), the symbol with the highest matching degree among the symbols (symbol candidate images) determined to be the same in the dictionary buffer 3005 is selected. A final symbol determination unit 3003-1 to be adopted as a final symbol, a symbol position information ID correction unit 3003-2 for correcting the ID of the symbol position information in the symbol position information buffer 3005-2, and a symbol information buffer 3003- 1 has an unnecessary symbol information deletion unit 3003-3 that deletes symbol information that has become unnecessary (not selected as a symbol).

  When the dictionary information output flag F3001 is ON, the dictionary information output unit 3004 displays the information (Symbol information I3004 and Symbol position information I3005) of the Symbol information buffer 3005-1 and Symbol position information buffer 3005-2 in the dictionary buffer 3005. A gate 3004- which controls to output and sends a signal for initializing the dictionary buffer 3005 (including the symbol information buffer 3005-1 and the symbol position information buffer 3005-2) to the dictionary buffer initialization unit 3004-2 after the output. 1 (when the dictionary information output flag F3001 is OFF, the gate does not perform the output / initialization process as described above) and the dictionary buffer 3005 (the Symbol information buffer 3005-1 and the Symbol position information buffer 3005-2 / matching Symbol information). A dictionary buffer initialization unit 3004-2 for initializing the buffer 3005-3.

  For example, assume that the image shown in FIG. 40 is input. Here, a description will be given of an example of selecting a dictionary that seems to be optimal when dictionary registration is completed.

  The operation of the lexical object data extraction unit 3000 will be described. First, the lexicon target data D3001 is extracted from the input image of FIG. 40 (ST3001). When black pixels are searched from left to right and from top to bottom, (1,1) black pixels are found (the upper left is (0,0)).

  By extracting the connected component of (1,1) black pixels, “T” at the left end of FIG. 40 can be obtained as a candidate (a known method is applied as a method for extracting data to be dictionaryd).

  The operation of the symbol coincidence determination unit 3001 will be described. The dictionary-targeted data D3001 obtained by the dictionary-targeted data extraction unit 3000 is compared with the Symbol existing in the dictionary buffer 3005 (ST3002). Since the symbol to be compared is not registered in the dictionary buffer 3005 this time, a mismatch result (match determination result R3002) is output (ST3003, NO). It is also possible to intentionally put a specific symbol in the dictionary buffer beforehand.

  The operation of the symbol / matching symbol information / symbol position information registration unit 3002 will be described. If the match determination result R3002 is “match” (ST3003, YES), the following operation is performed (ST3006).

Registration of data D3002 as a symbol in the dictionary buffer 3005 as a symbol Registration of information about which existing symbol matches with the dictionary buffer 3005 Information indicating the position of the symbol in the dictionary buffer 3005 If the coincidence determination result R3002 is “non-coincidence” (NO in ST3003), the following operation is performed (ST3005).

Registration of data D3001 as a symbol in the dictionary buffer 3005 Information indicating the position of the symbol is registered in the dictionary buffer 3005 In this embodiment 3-1, first, it is “mismatch” (NO in ST3003). ,
・ Register the image “T” at the left end as a symbol in the dictionary buffer 3005 (ST3005) ・ Register the information that the position of the “T” at the left end is (1, 1) in the dictionary buffer 3005 (ST3005)
This state is shown in FIG.

  Subsequently, the next dictionary conversion target data extraction unit 3000 is operated. That is, the lexicon target data D3001 is extracted from the input image (ST3001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  When searching for black pixels from left to right and from top to bottom, a black pixel of (10,1) is found (the upper left is (0,0): the leftmost “T” is already registered, so it is excluded and searched) The corresponding “T” may be deleted when the image is input to the extraction unit).

  By extracting the connected component of the black pixel of (10, 1), the second T from the left in FIG. 40 can be obtained as a candidate (a known method is applied as the method for extracting data to be dictionaryd).

The operation of the symbol coincidence determination unit 3001 will be described. The dictionary-targeted data D3001 obtained by the dictionary-targeted data extraction unit 3000 is compared with the Symbol existing in the dictionary buffer 3005 (ST3002). Since the symbol to be compared exists in the dictionary buffer 3005 this time, the comparison is performed. Since the leftmost “T” is registered in advance, it is compared with the second “T” from the left. Here, it is assumed that a match result (match determination result R3002) is output (ST3003, YES). As a determination method, a known method can be applied.
The operation of the symbol / matching symbol information / symbol position information registration unit 3002 will be described. In the present embodiment 3-1, since it is “match” (ST3003, YES),
・ The second image “T” from the left is registered as a symbol in the dictionary buffer 3005 (ST3004).
-Information that the second "T" from the left matches "T" at the left end is registered in the dictionary buffer 3005 (ST3004)
-Information that the position of the second "T" from the left is (10, 1) is registered in the dictionary buffer 3005 (ST3004)
This state is shown in FIG.

  Subsequently, the next dictionary conversion target data extraction unit 3000 is operated. That is, the lexicon target data D3001 is extracted from the input image (ST3001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  When searching for black pixels from left to right and from top to bottom, a black pixel of (19,1) is found (the upper left is (0,0): Exclude the left edge and the right “T” because it is already registered) Search: The corresponding “T” may be deleted when an image is input to the main extraction unit).

  By extracting the connected component of (19, 1) black pixels, I in FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be dictionaryd.

  The operation of the symbol coincidence determination unit 3001 will be described. The dictionary-targeted data D3001 obtained by the dictionary-targeted data extraction unit 3000 is compared with the Symbol existing in the dictionary buffer 3005 (ST3002). Since the symbol to be compared exists in the dictionary buffer 3005 this time, the comparison is performed. Compare the left edge with “T” on the right. Here, it is assumed that a result of mismatch (match judgment result R3002) is output (ST3003, NO). A known method can be applied as the determination method.

The operation of the symbol / matching symbol information / symbol position information registration unit 3002 will be described. In the present embodiment 3-1, since it is “mismatch” (ST3003, NO),
・ Register image “I” as a symbol in dictionary buffer 3005 (ST3005)
-Information that the position of "I" is at (19,1) is registered in the dictionary buffer 3005 (ST3005)
This state is shown in FIG.

  Subsequently, the next dictionary conversion target data extraction unit 3000 is operated. That is, the lexicon target data D3001 is extracted from the input image (ST3001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  If you search for black pixels from left to right and from top to bottom, you can find (22,1) black pixels (the upper left is (0,0): “T”, “T”, “I” from the left are excluded because they are already registered) To search: the corresponding data may be deleted when the image is input to the main extraction unit).

  By extracting the connected component of (22,1) black pixels, the rightmost T in FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be dictionaryd.

  The operation of the symbol coincidence determination unit 3001 will be described. The dictionary-targeted data D3001 obtained by the dictionary-targeted data extraction unit 3000 is compared with the Symbol existing in the dictionary buffer 3005 (ST3002). Since the symbol to be compared exists in the dictionary buffer 3005 this time, the comparison is performed.

  At this time, since “T” and “I” are registered in the dictionary buffer 3005, “T” on the left end and “T” on the right side of “T” are determined to be the same. Compare only with]. When there are a plurality of comparison targets determined to be the same, a method of comparing all of them is also acceptable.

  Here, the result of coincidence (coincidence judgment result R3002) is output (ST3003, YES). A known method can be applied as the determination method.

The operation of the symbol / matching symbol information / symbol position information registration unit 3002 will be described. In the present embodiment 3-1, since it is “match” (ST3003, YES),
・ Register the image “T” on the right as a symbol in the dictionary buffer 3005 (ST3004)
-Information that “T” at the right end matches “T” at the left end is registered in the dictionary buffer 3005 (ST3004).
-Information that the position of the rightmost "T" is at (22,1) is registered in the dictionary buffer 3005 (ST3004)
This state is shown in FIG.

  Since all the dictionary registrations that the user has planned have been completed (ST3006, YES), the symbol determination flag F3006 is turned on, and after final determination of the symbol (ST3007), the dictionary output flag F3001 is turned on and the symbol information I3004. And Symbol position information I3005 is output (ST3008).

  The Symbol determination unit 3003 selects the symbol with the smallest error among the symbols in the dictionary buffer that have the same symbol (see the determination example described later). The selected symbol is registered as a representative of the same symbol candidate, and the other same symbol is deleted. Along with that, the link with the position information is corrected.

  If none of the symbols are identified as the same symbol, the symbol that inevitably exists is the representative symbol.

  If two are judged to be the same symbol, the one with less noise is used. As another method, it is also possible to adopt the one registered earlier / the one with smooth edges.

  Next, an example of determination will be shown.

-Symbol of ID1 is as follows.

Symbol of ID1 and ID2 is 4 pixels different
Symbols of ID1 and ID4 are different by 7 pixels ⇒11 pixels in total are different ・ Symbols of ID2 are as follows.

ID2 and ID1 Symbols differ by 4 pixels
Symbols of ID2 and ID4 differ by 4 pixels ⇒Total of 8 pixels differ ・ Symbol of ID4 is as follows.

Symbol of ID4 and ID1 is 7 pixels different
Symbols of ID4 and ID2 differ by 4 pixels ⇒11 pixels in total After the final Symbol is determined, the ID of Symbol position information is corrected accordingly, and symbol information that is no longer needed is deleted. In this case, the symbol information of ID1 and ID4 in the above determination example is deleted.

  FIG. 45 shows the final dictionary buffer state.

  The operation of the dictionary information output unit 3004 will be described. Symbol information I3003 and symbol position information I3004 existing in the dictionary buffer 3005 are output.

  As described above, there is an advantage that there is a high probability that the image quality is improved as compared with a case where an image first listed as a dictionary candidate is registered as a dictionary (see Symbol) (see FIGS. 46 and 47).

  Also, as shown in the following 1-3, since the user can intentionally adjust the processing start timing of the Symbol determination unit, it is possible to create high-quality data while saving the dictionary buffer.

1. 1. When the specified dictionary buffer amount is reached 2. When the number of registered symbols (including provisional registration) reaches the specified number When the same number of symbols reaches the specified number (determines the symbol only for the specified symbol)
FIG. 49 is a diagram showing an example of the overall configuration of an image processing apparatus (image compression apparatus) according to Embodiment 3-2 of the present invention, and FIG. 48 is an image processing apparatus according to Embodiment 3-2 of the present invention. It is a figure which shows an example of this detailed structure.

  In the present embodiment 3-2, the symbol candidate image can be registered in the dictionary buffer 3005 only when designated by the user. A description will be given centering on differences from the embodiment 3-1. That is, the operation of the symbol / matching symbol information / symbol position information registration unit 3006 will be described. When the Symbol candidate image registration flag F3007 is ON, an ID is assigned to the lexical object data D3002, the lexical object data D3002 and its ID are registered in the Symbol information buffer 3005-1 of the dictionary buffer 3005, and the Symbol candidate image registration flag F3007. Is OFF, only when the match determination result R3003 does not match, an ID is assigned to the lexical object data D3002, and the lexical object data D3002 and its ID are registered in the symbol information buffer 3005-1 of the dictionary buffer 3005. An information / ID registration unit 3006-2, an ID / symbol position information registration unit 3006-2 for registering the ID and position information of the symbol (or symbol candidate image) to be registered in the symbol position information buffer 3005-2 of the dictionary buffer 3005; If the symbol candidate image registration flag F3007 is ON, the match determination result R3003 is other If it matches the Symbol (Symbol candidate image), with a matching Symbol information registration unit 3006-3 to register Symbol ID number match the match Symbol information buffer 3005-3 dictionary buffer 3005.

  The user turns ON the symbol candidate image registration flag F3007 when he / she wants to register the lexicographic object data D3002 as the symbol candidate image in the dictionary buffer 3005, and sets it OFF otherwise.

  When the symbol candidate image registration flag F3007 is ON, the symbol / matching symbol information / symbol position information registration unit 3006 operates in the same manner as the symbol / matching symbol information / symbol position information registration unit 3002, and stores the symbol candidate image in the dictionary buffer. Store up in 3005.

  When the symbol candidate image registration flag F3007 is OFF, the symbol / matching symbol information / symbol position information registration unit 3006 newly registers the symbol in the dictionary buffer 3005 only when the matching does not match. If they match, the ID and position information of the matching Symbol are registered in the dictionary buffer 3005. If they do not match, the ID of the newly registered Symbol and its position information are registered in the dictionary buffer 3005. Symbol candidate images are not accumulated.

  By having such a mechanism, it is possible to improve the image quality while suppressing the amount of memory used.

  For example, the user can limit the amount of memory used by performing the following processing and improve the image quality.

  (1) Symbol candidate images are stored until the amount of used memory reaches a specified value.

  (2) When the amount of memory used exceeds the specified value, a symbol is selected and unnecessary symbol candidate images are deleted.

  (3) Perform dictionary processing.

  The above-described Embodiments 3-1 and 3-2 are summarized below.

  (1) An image processing apparatus is an apparatus for compressing an image by forming one dictionary bitmap for each character and storing the dictionary bitmap and position information, and a group of bitmaps determined to be the same dictionary (A) Adopt the one with the least change point of white pixels and black pixels, (b) Adopt the one with the least noise pixels, (c) Adopt the one with the highest degree of coincidence by OCR determination The dictionary BMP is determined.

  More specifically, it is as follows.

(2) The image processing apparatus separates image information into a dictionary bitmap (Symbol) and its position information and stores the data to be compressed by inputting the image to be compressed, and outputs the data to be dictionaryd (dictionary candidate image). The extraction unit, the dictionary target data as input, the symbol match determination unit for determining whether or not it matches the symbol registered in the dictionary buffer, the dictionary target data and the match determination result as input, and the symbol is If they match, register the data to be dictionaryd (Symbol), the matching symbol information (ID, etc.), and the symbol position information in the dictionary buffer. If the symbols do not match, the data to be dictionaryd (Symbol ), Symbol position information in the dictionary buffer, a symbol / matching symbol information / symbol position information registration unit, and a dictionary information output unit that outputs the symbol information and symbol position information in the dictionary buffer. ,
After the image of the data to be dictionaryd is determined to match the symbol, it is saved in the dictionary buffer as a symbol candidate image, and the symbol candidate with the highest degree of pixel matching among the symbol candidate images determined to be the same symbol Register the image as a symbol.

  (3) The image processing apparatus separates image information into a dictionary bitmap (Symbol) and its position information, and stores the data to be compressed by inputting the image to be compressed and outputting the data to be dictionaryd (dictionary candidate image). The extraction unit, the dictionary target data as input, the symbol match determination unit for determining whether or not it matches the symbol registered in the dictionary buffer, the dictionary target data and the match determination result as input, and the symbol is If they match, register the data to be dictionaryd (Symbol), the matching symbol information (ID, etc.), and the symbol position information in the dictionary buffer. If the symbols do not match, the data to be dictionaryd (Symbol ), A symbol position information registration unit for registering the symbol position information in the dictionary buffer, and a dictionary information output unit for outputting the symbol information and the symbol position information of the dictionary buffer. At any timing, after the image of the data to be dictionaryd is determined to match the symbol, it is stored in the dictionary buffer as a symbol candidate image, and the most pixel among the symbol candidate images determined to be the same symbol Symbol candidate images with a high degree of coincidence can be registered as Symbols, or the same operation as a conventional compression processing apparatus can be performed.

  Next, a fourth embodiment of the present invention will be described.

  There is a method of compressing an image by forming a dictionary bitmap for each character or the like, compressing the dictionary bitmap, and storing the compressed image data and position information. However, the dictionary bitmap created by this method depends only on the shape of the input image, and the compression rate of the dictionary bitmap is not always good for a scanned image or the like.

  In more detail, if dictionary compression is applied to scanned images, etc., there is very little probability of 100% pixel matching even for the same character, so a dictionary that is considered to be the same to some extent is a dictionary that is considered the same dictionary (symbol) There is compression. Scanned images are likely to contain noise, etc., and may be present in Symbols registered as dictionaries. This becomes a factor of image quality degradation of the final output image.

  As a countermeasure, there is a method of having a clean font dictionary (symbol) in advance, and using that dictionary (symbol) if applicable, and vectorizing (quasi-reversible) otherwise (Japanese Patent Laid-Open No. 2005-208772). ). However, the vectorization of characters raises concerns about a decrease in compression rate.

  Therefore, the image processing apparatus (image compression apparatus) according to the fourth embodiment of the present invention can improve the image quality of the symbol by optimizing the symbol in a timely manner (edge sharpening / noise removal) and at the same time high compression. (See FIG. 50). That is, the image processing apparatus according to the fourth embodiment of the present invention increases the compression rate by changing the shape of the dictionary bitmap so that the compression rate is high (eg, connecting runs: giving the shape a feature). To do. This makes it possible to efficiently compress the dictionary bitmap.

  Specifically, as shown in FIGS. 51 and 52, the image processing apparatus according to the fourth embodiment of the present invention includes a lexicization target range determination unit that receives electronic data and outputs a lexicalization target range, A lexicon determination unit that outputs data and a lexicalization target range and outputs a lexicographic determination result for determining whether or not to use a dictionary bit map; an electronic data and a lexicographic determination result as inputs; A dictionary BMP creation unit that creates and outputs dictionary bitmap information in the case of a decision to create a dictionary bitmap, and inputs a lexicalization target range, lexicographic judgment result, and dictionary buffer (dictionary bitmap information group) as a dictionary A character position information creation unit that outputs character position information that associates the target dictionary bitmap information with its position, and converts the data of the dictionary bitmap information group in the dictionary buffer into a form that can be easily compressed. Comprising a dictionary bitmap shape conversion unit for force. The dictionary buffer is a buffer that collectively manages dictionary bitmap information output from the dictionary BMP creation unit.

  More specifically, the shape of the dictionary bitmap is changed in consideration of input data by adding input image / character position information to the input of the shape conversion unit. Further, the dictionary information generating means is controlled by input parameters. By adopting such a configuration, it is possible to compress the dictionary bitmap more efficiently.

  Hereinafter, Embodiment 4 of the present invention will be described in detail.

  51 is a diagram showing the concept of compression processing by an image processing apparatus (image compression apparatus) according to Embodiment 4-1 of the present invention, and FIG. 54 is an image processing apparatus according to Embodiment 4-1 of the present invention. FIG. 53 is a diagram illustrating an example of an overall configuration of an (image compression apparatus), and FIG. 53 is a diagram illustrating an example of a detailed configuration of an image processing apparatus (image compression apparatus) according to Embodiment 4-1 of the present invention. FIG. 55 is a diagram showing an example of the compression processing according to Embodiment 4-1 of the present invention. The image processing apparatus shown in FIG. 51 corresponds to the image processing apparatus shown in FIGS. 53 and 54, and mainly performs compression processing according to Embodiment 4-1 with reference to the image processing apparatus shown in FIGS. explain.

  The symbol match determination unit 4001 compares / determines whether the dictionary target data D4002 and the symbol information I4004 existing in the dictionary buffer 4005 match, and matches the determination result R4003 (if they do not match, it means that they do not match). It has a symbol comparison unit 4001-1 and a symbol comparison result output unit 4001-2 for outputting information (if matching, the ID number of the matching symbol).

  The Symbol coincidence determination unit 4001 is a known means / method, and 100% pixels do not necessarily have to coincide with each other, and includes a means / method for determining that something similar to some extent is the same.

  The Symbol / Symbol position information registration unit 4002 registers the data to be dictionaryd D4002 as new symbol information, assigns an ID number, and registers it in the Symbol information buffer 4005-1 of the dictionary buffer 4005. ID / Symbol position information registration unit for registering the position information in the image and the ID of the symbol in the symbol position information buffer 4005-2 of the dictionary buffer 4005 4002-2.

  If the match determination result R4003 is a result of determining that the dictionary target data D4002 and the symbol information do not match, after the processing of the symbol information / ID registration unit 4002-1, the ID / symbol position is determined. If the match determination result R4003 is a result of determining that the data to be dictionary-matched D4002 and the symbol of the ID with the symbol match, as in the processing of the information registration unit 4002-2, the ID / Symbol position information It has a selector that distributes the processing of the registration unit 4002-2.

  In order to improve the image quality of the symbol in the dictionary buffer 4005, the symbol correction unit 4003 includes an edge sharpening unit 4003-1 that smoothes the unevenness of the edge, and a noise removal unit 4003-2 that removes noise such as noise.

  This correction process is performed on a newly registered Symbol every time a new Symbol is registered in the dictionary buffer 4005.

  When the dictionary information output flag F4001 is ON, the dictionary information output unit 4004 outputs the information (Symbol information I4004 and Symbol position information I4005) of the Symbol information buffer 4005-1 and Symbol position information buffer 4005-2 in the dictionary buffer 4005. The gate 4004-1 sends a signal for initializing the dictionary buffer 4005 (including the symbol information buffer 4005-1 and the symbol position information buffer 4005-2) to the dictionary buffer initialization unit 4004-2 after output. (When the dictionary information output flag F4001 is OFF, the gate does not perform the above output / initialization processing) and the dictionary buffer 4005 (including the symbol information buffer 4005-1 and the symbol position information buffer 4005-2) is initialized. And a dictionary buffer initialization unit 4003-2 to be converted.

  In the present embodiment 4-1, a case where the user controls the input signal as follows will be described.

The image in FIG. 40 described above is set as an input image.

The dictionary information output flag F4001 is always OFF except for the last page, and is turned ON when all dictionary registrations for the last page are completed. When all the processes are completed, Symbol information I4004 and Symbol position information I4005 are output.

  Next, an example of dictionary processing will be described. The operation of the lexical object data extraction unit 4000 will be described. First, the lexicon target data D4002 is extracted from the input image of the first page in FIG. 40 (ST4001).

  When black pixels are searched from left to right and from top to bottom, (1,1) black pixels are found (the upper left is (0,0)). By extracting the connected component of the black pixel of (1, 1), “T” at the left end of the image of the first page in FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be lexicized.

  The operation of the symbol coincidence determination unit 4001 will be described. The lexicalization target data D4002 obtained by the lexicalization target data extraction unit 4000 is compared with the Symbol existing in the dictionary buffer 4005 (ST4002). Since the symbol to be compared is not registered in the dictionary buffer 4005 this time, a mismatch result (match determination result R4003) is output (ST4003, NO). It is also possible to intentionally put a specific symbol in the dictionary buffer 4005 in advance.

  The operation of the Symbol / Symbol position information registration unit 4002 will be described. If the match determination result R4003 is “match (corresponding ID number)” (ST4003, YES), the following operation is performed (ST4004).

-Information (ID) that identifies which symbol information matches is registered in the dictionary buffer 4005 (ST4004)
-Information indicating the position of the symbol is registered in the dictionary buffer 4005 (ST4004).
If coincidence determination result R4003 is “non-coincidence” (NO in ST4003), the following operation is performed (ST4005).

・ Registered data 1003 as a symbol is registered in the dictionary buffer 4005 as a symbol (ST4005).
-Information indicating where the symbol is located is registered in the dictionary buffer 4005 (ST4005).
In this embodiment, since it is “mismatch” (ST4003, NO), “T” at the left end of the first page is registered as a symbol in the dictionary buffer 4005 (ST4005) • “T” is (1,1) of the first page ) Is registered in the dictionary buffer 4005 (ST4005).
The state of the dictionary buffer 4005 at this time is shown in FIG.

  When a new symbol is registered in the dictionary buffer 4005, the symbol correction unit 4003 performs processing for smoothing an edge (edge sharpening processing 4003-1) and noise removal such as noise (noise removal). Part 4003-2) (ST4006).

  The state of the dictionary buffer 4005 at this time is shown in FIG.

  Subsequently, the next dictionary conversion target data extraction unit 4000 is operated. The dictionary data D4002 is extracted from the input image of the first page in FIG. 40 (ST4001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  When searching for black pixels from left to right and from top to bottom, a black pixel of (10,1) is found (the upper left is (0,0): the leftmost “T” is already registered, so it is excluded and searched) It is also possible to delete the corresponding “T” when inputting the image to the extraction unit).

  By extracting the connected component of (10, 1) black pixels, the second “T” from the left in FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be dictionaryd.

  The operation of the symbol coincidence determination unit 4001 will be described. The lexicalization target data D4002 obtained by the lexicalization target data extraction unit 4000 is compared with the Symbol existing in the dictionary buffer 4005 (ST4002). Since the symbol to be compared is registered in the dictionary buffer 4005 this time, the comparison is performed. Compare with “T” on the left. Here, a result of matching with “T” at the left end (matching determination result R4003) is output (ST4003, YES). A known method can be applied as the determination method.

The operation of the Symbol / Symbol position information registration unit 4002 will be described. “Matches the leftmost“ T ””, so
・ Register information (ID number) in which the matching Symbol is identified as “T” at the left end in the dictionary buffer 4005 (ST4004)
・ Register location information that this "T" is at (10,1) in the dictionary buffer 4005 (ST4004)
The state of the dictionary buffer 4005 at this time is shown in FIG.

  The Symbol correction unit 4003 does not perform any particular processing because there is no newly registered Symbol this time.

  Subsequently, the next dictionary conversion target data extraction unit 4000 is operated. The dictionary data D4002 is extracted from the input image on the second page of FIG. 40 (ST4001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  If you search for black pixels from left to right and from top to bottom, you can find (19,1) black pixels (the upper left is (0,0): the two left “T” are already registered, so search for them: It is also possible to delete the corresponding “T” when inputting an image into the main extraction unit).

  By extracting the connected component of the black pixel of (19, 1), “I” in FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be dictionaryd.

  The operation of the symbol coincidence determination unit 4001 will be described. The lexicalization target data D4002 obtained by the lexicalization target data extraction unit 4000 is compared with the Symbol existing in the dictionary buffer 4005 (ST4002). Since the symbol to be compared is registered in the dictionary buffer 4005 this time, the comparison is performed. Compare with “T”. Here, a result of mismatch (match judgment result R4003) is output (ST4003, NO). A known method can be applied as the determination method.

The operation of the Symbol / Symbol position information registration unit 4002 will be described. Because it is “mismatch” (ST4003, NO),
・ "I" is registered as a symbol in the dictionary buffer 4005 (ST4005)
-Information that "I" is at the position (19, 1) is registered in the dictionary buffer 4005 (ST4005).
The state of the dictionary buffer 4005 at this time is shown in FIG.

  When a new symbol is registered in the dictionary buffer 4005, the symbol correction unit 4003 performs processing for smoothing an edge (edge sharpening processing 4003-1) and noise removal such as noise (noise removal). Part 4003-2) (ST4006).

  The state of the dictionary buffer 4005 at this time is shown in FIG.

  Subsequently, the next dictionary conversion target data extraction unit 4000 is operated. The lexical object data extraction unit 4000 extracts lexical object data D4002 from the input image on the second page of FIG. 40 (ST4001). Here, extraction is performed by excluding Symbols registered and determined earlier.

  If you search for black pixels from left to right and from top to bottom, you will find (19,1) black pixels (the upper left is set to (0,0): the left two “T” and “I” are already registered, so exclude them) Search: The corresponding “T” and “I” can be deleted when the image is input to the main extraction unit).

  By extracting the connected component of the black pixel of (22, 1), “T” at the right end of FIG. 40 can be obtained as a candidate. A known method can be applied as a method for extracting data to be dictionaryd.

  The operation of the symbol coincidence determination unit 4001 will be described. The lexicalization target data D4002 obtained by the lexicalization target data extraction unit 4000 is compared with the Symbol existing in the dictionary buffer 4005 (ST4002). Since the symbol to be compared is registered in the dictionary buffer 4005 this time, the comparison is performed. Compare with “T” and “I”. Here, a result of matching with “T” at the left end (matching determination result R4003) is output (ST4003, YES). A known method can be applied as the determination method.

The operation of the Symbol / Symbol position information registration unit 4002 will be described. Since it is “matches“ T ”at the left end” (ST4003, YES),
・ Register the information (ID number) in which the matching Symbol is recognized as “T” at the left end in the dictionary buffer 4005 (ST4004)
-Location information that this "T" is at (10,1) on the second page is registered in the dictionary buffer 4005 (ST4004).
The state of the dictionary buffer 4005 at this time is shown in FIG.

  The Symbol correction unit 4003 does not perform any particular processing because there is no newly registered Symbol this time.

  Since all the registered dictionary registrations are completed (YES in ST4007), the user turns on dictionary information output flag F4001 and takes out symbol information I4004 and symbol position information I4005.

  Since the dictionary information output flag F4001 is ON, the dictionary information output unit 4003 outputs the symbol information I4004 and the symbol position information I4005 in the dictionary buffer 4005 (ST4008), and then initializes the dictionary buffer 4005 (ST4009).

  As described above, there is an advantage that the image quality is improved as compared with a case where an image initially listed as a dictionary candidate is directly registered as a dictionary. When an image initially listed as a dictionary candidate is directly registered as a dictionary, it becomes a decoded image as shown in FIG. On the other hand, the decoding result of the embodiment 4-1 is as shown in FIG.

  FIG. 64 is a diagram showing an example of the overall configuration of an image processing apparatus (image compression apparatus) according to Embodiment 4-2 of the present invention, and FIG. 63 is an image processing apparatus according to Embodiment 4-2 of the present invention. It is a figure which shows an example of a detailed structure of (image compression apparatus). FIG. 65 is a diagram illustrating an example of compression processing according to Embodiment 4-2 of the present invention.

  Symbol correction section 4006 is a processing section in which symbol edge sharpening processing and noise removal processing are operated (ST4006b) at the symbol correction timing designated by the user (ST4006a, YES). Thus, it is possible to prevent a phenomenon such as “a matching dictionary candidate image is changed from a conventional one that does not correct the symbol”, which is caused by immediately correcting the registered Symbol. There is an advantage that the symbol can be corrected at the timing when the user wants to correct it.

  The above Embodiments 4-1 and 4-2 will be summarized below.

  (1) The image processing apparatus receives electronic data and outputs a lexicographic target range determination unit, and determines whether or not the electronic data and the lexicographic target range are input to form a dictionary bitmap. A dictionary determining unit for outputting a dictionary determination result to be generated, and a dictionary for inputting and outputting electronic data and the dictionary determination result, and generating and outputting dictionary bitmap information when the dictionary determination result is a determination to create a dictionary bitmap Character that outputs character position information that associates the dictionary bitmap information to be lexicized and its position, with the BMP creation unit, lexicalization target range, lexicographic judgment result, and dictionary buffer (dictionary bitmap information group) as inputs A position information creation unit and a dictionary bitmap shape conversion unit for converting the data of the dictionary bitmap information group of the dictionary buffer into a form that can be easily compressed and outputting the data. Further, by adding the input image / character position information to the input of the shape conversion unit, the shape of the dictionary bitmap is changed in consideration of the input data. Further, the dictionary information generating means is controlled by the input parameters.

  More specifically, it is as follows.

  (2) The image processing apparatus separates image information into a dictionary bitmap (Symbol) and its position information and stores the data to be compressed by inputting the image to be compressed, and outputs the data to be dictionaryd (dictionary candidate image). The extraction unit, the dictionary target data as input, the symbol match determination unit for determining whether or not it matches the symbol registered in the dictionary buffer, the dictionary target data and the match determination result as input, and the symbol is If they match, register the data to be dictionaryd (Symbol), the matching symbol information (ID, etc.), and the symbol position information in the dictionary buffer. If the symbols do not match, the data to be dictionaryd (Symbol ) And a symbol / symbol position information registration unit for registering the symbol position information in the dictionary buffer, and a dictionary information output unit for outputting the symbol information and the symbol position information of the dictionary buffer. A process for smoothing and noise removal processing.

  (3) The image processing apparatus separates image information into a dictionary bitmap (Symbol) and its position information, and stores the data to be compressed by inputting the image to be compressed and outputting the data to be dictionaryd (dictionary candidate image). The extraction unit, the dictionary target data as input, the symbol match determination unit for determining whether or not it matches the symbol registered in the dictionary buffer, the dictionary target data and the match determination result as input, and the symbol is If they match, register the data to be dictionaryd (Symbol), the matching symbol information (ID, etc.), and the symbol position information in the dictionary buffer. If the symbols do not match, the data to be dictionaryd (Symbol ) And a symbol / symbol position information registration unit for registering the symbol position information in the dictionary buffer, and a dictionary information output unit for outputting the symbol information and the symbol position information of the dictionary buffer. A process of smoothing, performed at the timing designated by the user noise removal processing.

  Next, a fifth embodiment of the present invention will be described.

  A technology that compresses images by comparing images and comparing whether they are the same or by examining whether the same image as a certain first image exists in the second image. Proposed.

  ISO / IEC14492 discloses the technology of JBIG2, a binary compression technology that has become an international standard. This treats a certain region or character (character string, etc.) of an image as a single dictionary image (for example, a single character image), treats common items as the same dictionary image, and compresses data by having the dictionary image and position information. It is a method to do. These are effective for images having a specific pattern (such as character images and halftone images). Applying this to the scanned image, matching with the dictionary image while searching for a specific pattern (image), and if there is no match, registering it in the dictionary makes it possible to achieve high compression of the scanned image.

  Japanese Patent Application Laid-Open No. 2006-23976 discloses a technique for examining whether an image is the same as or similar to an arbitrary template image. The basic technique of Japanese Patent Laid-Open No. 2006-23976 is called a template matching method, and in order to check whether an image is the same as a template image (first image) in the image, the first image is added to the second image. This is a technique for matching (comparison) while shifting little by little. As a method for reducing the amount of calculation while maintaining the matching accuracy of the template matching, a proposal as disclosed in JP-A-2006-23976 has been proposed. This technology uses template noise amount and template feature amount in template matching to determine “shift amount” when scanning while shifting the template, “deformation amount” considering image deformation, and whether image extraction is possible. Three parameters “threshold” are automatically set. In these, three parameters are automatically set in accordance with the template characteristics, and matching according to the template can be determined.

  The above two methods are similar methods for comparing images. However, a technique for comparing and creating a dictionary by searching for and creating images to be compared and images that are similar to each other in advance are similar images. It is very different in terms of technology to compare whether there is any.

  In ISO / IEC14492, by using a processing device as shown in FIG. 82, a feature image (for example, a character) is extracted from an input image, the same character image is held as one dictionary image, and has position information corresponding thereto. Therefore, it is possible to compress the data by reducing the data amount. However, in the processing apparatus as shown in FIG. 82, since it is not possible to make a coincidence determination (matching parameters are set for each feature image) by capturing the features for each feature image, there is a high possibility of erroneous determination.

  Japanese Patent Application Laid-Open No. 2006-23976 discloses a technique related to improving the accuracy and speeding up of template matching. As shown in FIG. 83, the technique disclosed in Japanese Patent Application Laid-Open No. 2006-23976 determines a “threshold value” that determines whether image extraction is possible, a “shift amount” of the template image, and a “deformation amount” of the image. However, since this is uniquely determined for the template image and does not use the characteristics of the image to be searched, there may be no effect depending on the target image. This technique determines the above three parameters using template information, but this is only applicable to template matching, and extracts the same feature image (for example, characters) from the input image, such as ISO / IEC14492, and dictionary Since it differs depending on the configuration when creating an image, it cannot be applied (in the first place, there is no concept such as shift amount like template matching in the apparatus as shown in FIG. 82, and there is no comparison purpose or configuration for realization at all. Different).

  The image processing apparatus (image compression apparatus) according to the fifth embodiment performs matching with a dictionary image while searching for a specific pattern (image) from binary or multi-valued input images, and if there is no match, registers it in the dictionary. The apparatus is excellent in reducing erroneous determination of image matching. The parameter for determining the match between the Symbol (Pattern) and the dictionary image, which is determined when the dictionary is dynamically created, is automatically determined using the information of the Symbol (Pattern) image to be compared with the dictionary image.

  The image processing apparatus according to the fifth embodiment extracts a certain area from an arbitrary input image, determines whether the area image matches one or more dictionary images, and registers them in the dictionary if they do not match. The image processing apparatus according to the fifth embodiment includes the following units.

・ Area image extraction unit ・ Matching parameter determination unit ・ Match determination unit ・ Dictionary registration unit (・ Dictionary)
With the above configuration, the image processing apparatus according to the fifth embodiment can perform matching determination with an optimal matching parameter for each input region image, and can create a dictionary while reducing erroneous determination.

  A case will be described in which characters are extracted from the input image shown in FIG. 66, matching between the characters and the dictionary is determined, and a minimum dictionary image is to be created.

  When the image processing apparatus according to the fifth embodiment is not applied, that is, in the case of the lossy parameter (those with a loose match determination condition), the determination is as follows.

・ A at both ends is determined to be coincident. L and 1 at the center are also coincident. The final image is as shown in FIG. 67, making it impossible to distinguish between “l” and “1”.

  Further, in the case of a near-lossless parameter (one with strict match determination conditions), the determination is made as follows.

・ A is determined to be inconsistent at both ends. ・ The central l and 1 are also determined to be inconsistent. If the parameters are stricter to distinguish “l” and “1”, the final image will be as shown in FIG. A ”also does not match, and high compression is not possible.

  The image processing apparatus according to the fifth embodiment processes a comparison of a large symbol with a lossy parameter and a comparison of a small symbol with a near-lossless parameter. As a result, the determination is made as follows.

• A at both ends is determined to be coincident • 1 and 1 at the center are also determined to be inconsistent. That is, by adding the above conditions, it is possible to effectively compress while maintaining the required image quality.

  Hereinafter, Embodiment 5 of the present invention will be described in detail.

  FIG. 70 is a diagram showing an example of the overall configuration of an image processing apparatus (image compression apparatus) according to Embodiment 5-1 of the present invention, and FIG. 76 is an image processing apparatus according to Embodiment 5-1 of the present invention. It is a figure which shows the detail of the matching parameter determination part 5001 of (image compression apparatus).

  The image processing apparatus according to Embodiment 5-1 extracts a feature image (character image) from a binary or multi-valued input image, and characters that appear to be the same exist in the input image by extracting them as the same dictionary image. Extract the character type. The feature image refers to all images having features that can be repeatedly generated, such as symbols and textures, in addition to the character image.

  The image processing apparatus according to Embodiment 5-1 can obtain a binary or multi-value character image that is not duplicated and included in the input image by inputting a binary or multi-value input image. Thereby, it is possible to confirm what type of character is present in the binary or multi-valued input image. This can be used, for example, to check whether emergency kanji (special characters) are used.

  71 shows an example of an input image, and FIGS. 72 to 75 are diagrams showing an example of an input image buffered in the dictionary buffer 5003. FIG. For example, the feature image extraction unit 5000 sequentially extracts characters from the input image shown in FIG. 71 using existing character segmentation processing, and outputs the feature images (feature image information I5001) shown in FIGS. .

  FIG. 76 is a block diagram illustrating a schematic configuration of a matching parameter determination unit 5001 that determines the above-described matching parameters. The matching parameter determination unit 5001 receives the feature image information I5001 extracted from the feature image extraction unit 5000 and determines a matching parameter. Here, the matching parameter is a parameter for controlling ON / OFF of a threshold value for determining whether or not images match or does not match, and operation of a subfunction of the matching algorithm.

  These parameters are determined from the feature image information I5001. As a method for determining this parameter, it is determined using the image width / height of the feature image information I5001. The matching parameter determination unit 5001 indicates a condition such as a strict matching condition when “image width × 2 <image height”, and determines a matching parameter.

  For example, when “image width × 2 <image height”, the matching parameter is adjusted so that the matching condition becomes strict, and when “image width × 2> = image height”, the matching parameter is set so that the matching condition becomes loose. Adjust.

  The feature image “A” at the left end of the input image shown in FIG. 72 satisfies the latter condition, so a parameter with a loose matching condition is selected. The matching parameter P5003 determined here is output.

  Examples of other matching parameter determination methods include the following.

  (1) Extract metadata (image width / height / resolution / annotation information, etc.) of the collected image information, and determine matching parameters using the information.

Example 1: Using image width
Characters with a narrow width such as f or l are likely to be erroneously determined. Therefore, it is possible to reduce the erroneous determination by tightening the matching condition only for characters with a narrow width.

Example 2: Annotation (character type: Japanese / English / numeric)
Matching errors between languages can be reduced. 1 (number 1) and l (English lowercase letter L) can also be clearly separated.

  (2) The amount of noise in the feature image information is extracted, and parameters are determined using the information. For example, a simple binary image of a binary image and an error diffusion character can be clearly distinguished.

  (3) The number of colors of the feature image information is counted, and the matching parameter is determined using the information. By looking at the color information in advance, the color matching conditions can be changed, and matching accuracy can be improved and processing speed can be increased.

  (4) The line width of the feature image information is extracted, and the matching parameter is determined using the information. Even if the thin line is a little thicker, it is conspicuous, so it is effective as a countermeasure. If all the matching judgments are made strict, the matching judgment rate for a thick line whose change in line width is not known is deteriorated.

  (5) The inclination of the feature image information is extracted, and the matching parameter is determined using the information. In the case of sentences etc., even if it is slightly tilted, it stands out, so it is effective as a countermeasure.

  The match determination unit 5002 receives the feature image information I5001 and the matching parameter P5002 determined by the matching parameter determination unit 5001, and determines whether the feature image information I5001 matches (similar) the dictionary image registered in the dictionary buffer 5003. judge. The result (match / mismatch) is output as a match determination flag F5004.

  For example, since nothing is registered in the dictionary at this time, the match determination flag F5004 is output as a mismatch.

  If the match determination flag F5004 does not match, the dictionary registration unit 5004 registers the feature image information I5001 in the dictionary buffer 5003. If the match determination flag F5004 is matched, the feature image information I5001 is not registered in the dictionary buffer 5003.

  In the case shown in FIG. 72, the match determination flag F5004 does not match, so the feature image information I5001 is registered in the dictionary buffer 5003. Similar processing is performed for the following “1”, “l”, and “A”, and FIGS. 72 to 75 show examples in which only A at both ends coincides.

  When the matching judgment condition (matching condition) is fixed, for example, when the matching condition is uniformly strict, all the characters of the input image shown in FIG. The difference in the level of characters you want to do is considered to be inconsistent.

  On the contrary, if the matching condition is loosened uniformly, “A” at both ends is determined to be coincident, but “1” and “l” at the center are also determined to be the same, I can't get the information I want at the time of restoration.

  The image processing apparatus according to the embodiment 5-1 can make the matching condition strict only for feature images that are likely to be mistaken in advance by dynamically changing the matching parameters. .

  FIG. 77 is a diagram illustrating an example of the overall configuration of an image processing apparatus (image compression apparatus) according to Embodiment 5-2 of the present invention.

  The image processing apparatus according to the embodiment 5-2 extracts a feature image (character image) from a binary or multi-valued input image, stores characters that are considered to be the same as the same dictionary image, and further stores the position information thereof. By saving, the amount of data required can be reduced. A feature image refers to all images having features that can be repeatedly generated, such as symbols and textures, in addition to character images.

  The image processing apparatus according to the embodiment 5-2 inputs a binary or multi-valued input image, so that a binary or multi-value character image included in the input image is present in each input image. Position information can be obtained. By obtaining such information, it is possible to achieve high image compression using JBIG2 or the like of SymbolDictionary of ISO / IEC14492.

  71 shows an example of the input image, and FIGS. 78 to 81 are diagrams showing an example of the input image buffered in the dictionary buffer 5003 and the position information buffered in the position information buffer 5005. For example, the feature image extraction unit 5000 sequentially extracts characters from the input image shown in FIG. 71 using existing character segmentation processing, and outputs the feature images (feature image information I5001) shown in FIGS. . First, the left character image “A” (feature image information I5001) is extracted, and its position information (feature image position information I5002) is also output.

  The matching parameter determination unit 5001 receives the feature image information I5001 extracted from the feature image extraction unit 5000 and determines a matching parameter. Here, the matching parameter is a parameter for controlling ON / OFF of a threshold value for determining whether or not images match or does not match, and operation of a subfunction of the matching algorithm.

  These parameters are determined from the feature image information I5001. As a method for determining this parameter, it is determined using the image width / height of the feature image information I5001. The matching parameter determination unit 5001 indicates a condition such as a strict matching condition when “image width × 2 <image height”, and determines a matching parameter.

  For example, when “image width × 2 <image height”, the matching parameter is adjusted so that the matching condition becomes strict, and when “image width × 2> = image height”, the matching parameter is set so that the matching condition becomes loose. Adjust.

  A matching parameter determination unit 5001 that determines the above-described matching parameters is as shown in FIG.

  Since the feature image “A” at the left end of the input image shown in FIG. 78 satisfies the latter condition, a parameter with a loose matching condition is selected. The matching parameter P5003 determined here is output.

  Examples of other matching parameter determination methods include the following.

  (1) Extract metadata (image width / height / resolution / annotation information, etc.) of the collected image information, and determine matching parameters using the information.

Example 1: Using image width
Characters with a narrow width such as f or l are likely to be erroneously determined. Therefore, it is possible to reduce the erroneous determination by tightening the matching condition only for characters with a narrow width.

Example 2: Annotation (character type: Japanese / English / numeric)
Matching errors between languages can be reduced. 1 (number 1) and l (English lowercase letter L) can also be clearly separated.

  (2) The amount of noise in the feature image information is extracted, and parameters are determined using the information. For example, a simple binary image of a binary image and an error diffusion character can be clearly distinguished.

  (3) The number of colors of the feature image information is counted, and the matching parameter is determined using the information. By looking at the color information in advance, the color matching conditions can be changed, and matching accuracy can be improved and processing speed can be increased.

  (4) The line width of the feature image information is extracted, and the matching parameter is determined using the information. Even if the thin line is a little thicker, it is conspicuous, so it is effective as a countermeasure. If all the matching judgments are made strict, the matching judgment rate for a thick line whose change in line width is not known is deteriorated.

  (5) The inclination of the feature image information is extracted, and the matching parameter is determined using the information. In the case of sentences etc., even if it is slightly tilted, it stands out, so it is effective as a countermeasure.

  The match determination unit 5002 receives the feature image information I5001 and the matching parameter P5002 determined by the matching parameter determination unit 5001, and determines whether the feature image information I5001 matches (similar) the dictionary image registered in the dictionary buffer 5003. judge. The result (match / mismatch) is output as a match determination flag F5004.

  If they match, the matched dictionary information (ID etc .: matched dictionary information I5005) is also output.

  For example, since nothing is registered in the dictionary at this time, the match determination flag F5004 is output as a mismatch.

  If the match determination flag F5004 does not match, the dictionary registration unit 5004 adds management information such as an ID number to the feature image information I5001 and registers it in the dictionary buffer 5003. Also, the registration determination flag F5006 is turned ON, and information that registration is performed is output. The registered dictionary information (ID etc .: registered dictionary information I5007) is also output. When the coincidence determination flag F5004 is coincident, registration in the dictionary buffer 5003 is not performed, and output of the registered dictionary information I5006 is not necessary.

  In the case shown in FIG. 78, since the match determination flag F5004 does not match, management information “ID number = 1” is assigned to the feature image information I5001 and registered in the dictionary buffer 5003. The registered dictionary information (“ID = 1”: registered dictionary information I5007) is also output. The position information registration unit 5005 registers the position information of the feature image in the position information buffer 5006.

  If the match determination flag F5004 is matched, the match dictionary information I5005 and the feature image position information I5002 are linked and registered in the position information buffer 5006. If the match determination flag F5004 does not match, the registered dictionary information I5006 and the feature image position information I5002 are linked and registered in the position information buffer 5006.

  As shown in FIG. 78, since there is no coincidence at this point (left “A”), information such as ID1: coordinate (0,0) is stored in the position information buffer 5006. FIG. 78 shows the state of dictionary buffer 5003 and position information buffer 5006 at this time.

  Similar processing is performed for the next “1”, “l”, and “A”, and FIGS. 78 to 81 show examples in which only A at both ends coincides.

  If the parameters are fixed, for example, if the matching conditions are uniform and strict, the characters in the input image in FIG. 5 are all judged to be inconsistent, so the data amount cannot be reduced.

  On the contrary, if the matching condition is loosened uniformly, “A” at both ends is determined to be coincident, but “1” and “l” at the center are also determined to be the same, I can't get the information I want at the time of restoration.

  In the image processing apparatus according to the embodiment 5-2, by dynamically changing the matching parameter, it is possible to tighten the matching condition only for the feature image that is likely to be mistaken in advance. . Therefore, it is possible to reduce the amount of data while maintaining the image quality (information) (A at both ends coincides and two characters at the center do not coincide (result as shown in FIG. 71)).

  The above-described Embodiments 5-1 and 5-2 are summarized below.

  (1) The image processing apparatus extracts an area from an arbitrary input image, determines whether or not the area image matches one or more dictionary images, and registers the image in the dictionary if they do not match A device that generates one or more region images from an input image, a matching parameter determination unit that determines and outputs a matching parameter by using the region image created by the region image creation unit, and a region; The match determination unit that determines whether the region image created by the image creation unit matches the dictionary image in the dictionary using the matching parameter determined by the matching parameter determination unit and the match determination result determined by the match determination unit do not match A dictionary registration unit for registering the region image in the dictionary.

  More specifically, it is as follows.

  (2) The image processing apparatus determines whether the feature image and the dictionary image registered in the dictionary match by using the feature image extraction unit that extracts the feature image from the input image and the feature image and the matching parameter as input. A match determination unit that outputs a match determination result, and an image processing apparatus that inputs the match determination result and the feature image and registers the feature image information in the dictionary only when the match determination result does not match. A matching parameter determining unit for determining a matching parameter. Since matching can be performed in accordance with the feature of the feature image, highly accurate matching (matching determination) is possible. Therefore, a highly accurate dictionary image can be obtained.

  (3) An image processing apparatus extracts a feature image from an image and outputs feature image information and position information of the feature image; a feature image and a matching parameter as inputs; A match determination unit that outputs information on the dictionary that matches the match determination result, and a match determination result that is output from the feature image and the match determination unit. A dictionary registration unit for registering a feature image in the dictionary and outputting information of the registered dictionary, and a match dictionary that is output only when the match information is determined by the match determination result and the match determination unit. Information and registration dictionary information output from the dictionary registration unit only when there is a mismatch, and if the match determination result is a match, the matching dictionary information and the feature image position information are linked to Output, and if it does not match, the registered dictionary information and the feature image position information are linked and output as the position information of the dictionary, and the image processing apparatus includes a feature image as an input to determine a matching parameter. A parameter determination unit is provided. Since matching can be performed in accordance with the feature of the feature image, highly accurate matching (matching determination) is possible. Also, since the dictionary image and its position information are output, a highly accurate and highly compressed dictionary compressed file with few image quality defects can be generated.

  Next, a sixth embodiment of the present invention will be described.

  In order to efficiently compress a binary image, ISO / IEC14492 JBIG2 compression is a standardized method of converting an image bitmap into a dictionary and using an index (corresponding dictionary number and its arrangement) and a dictionary as compressed data. JBIG2 is classified into character area, halftone area, and other areas, and a compression method suitable for each is applied.

  However, in such a method that compresses by saving the symbol and its position information (dictionary compression), if it is judged that the difference is the same to some extent and the symbol is replaced, if the matching is missed, the information such as characters will be different in the worst case (Information changes).

  Therefore, the image processing apparatus (image compression apparatus) according to the sixth embodiment creates and decodes a complementary image that compensates for image quality degradation caused by irreversible matching, apart from dictionary data, as shown in FIGS. Sometimes image quality deterioration can be reduced by superimposing.

  FIG. 84 is a diagram showing an example of the overall configuration of an image processing apparatus (image compression apparatus) according to Embodiment 6 of the present invention. The image processing apparatus includes a matching determination unit 6000, a dictionary 6001, a selector 6002, a difference pixel extraction unit 6003, a difference pixel registration unit 6004, and a difference image buffer 6005.

  Specifically, as shown in FIG. 85, in dictionary compression by irreversible matching by the matching determination unit 6000 and the dictionary 6001, the difference pixel extraction unit 6003 extracts pixels in which black pixels change to white pixels, and registers the difference pixels. The unit 6004 has the extracted pixel group as another image, the difference image buffer 6005 outputs the difference image at the time of decoding, and prevents the pixel change by superimposing the difference image and the dictionary data. Note that black pixels: bit = 1 and white pixels: bit = 0.

  Also, as shown in FIG. 86, in dictionary compression by irreversible matching by the matching determination unit 6000 and the dictionary 6001, the difference pixel extraction unit 6003 extracts pixels in which white pixels change to black pixels, and the difference pixel registration unit 6004 Having the extracted pixel group as another image, the difference image buffer 6005 outputs the difference image at the time of decoding, and the pixel change is prevented by superimposing the difference image and the dictionary data. Note that black pixels: bit = 1 and white pixels: bit = 0.

  Also, as shown in FIG. 87, in dictionary compression by irreversible matching by the matching determination unit 6000 and the dictionary 6001, the difference pixel extraction unit 6003 extracts pixels whose information changes, and the difference pixel registration unit 6004 selects the extracted pixel group. The difference image buffer 6005 outputs the difference image at the time of decoding and prevents the pixel change by superimposing the difference image and the dictionary data. Note that black pixels: bit = 1 and white pixels: bit = 0.

  The above-described Embodiment 6 is summarized below.

  (1) The image processing apparatus stores the symbol and its position information, and when compressing the image, stores the difference in pixels, which occurs when the symbol is replaced and the symbol is replaced with the difference, as a difference image. As a result, compression can be performed effectively, and information deterioration due to dictionary formation (irreversible matching) is reduced. If there is no concern about a slight deterioration in image quality depending on the situation, only dictionary information can be decoded and high-speed display (no difference image can be decoded) is possible. In this way, high-speed display / image quality priority display can be switched.

  More specifically, it is as follows.

  (2) The image processing apparatus receives a compressed image by separating and storing the image information into a dictionary bitmap and its position information, and outputs a dictionary extraction target image (Symbol image), and a Symbol image Is input, and it is determined whether it matches the dictionary symbol registered in the dictionary, the symbol match determination unit that outputs the match determination result, the symbol image and the match determination result are input, and the symbol matches A symbol / symbol position information registration unit for registering the symbol information and the symbol position information in the dictionary if the symbols do not match. By inputting binary symbol image information and dictionary data, a pixel is determined by replacing the binary symbol with a symbol registered in the dictionary, and a matching determination unit 6000 that determines whether the symbol and the dictionary data match. Whether or not to process the difference pixel extraction unit 6003 that extracts pixels whose information changes, the difference pixel registration unit 6004 that registers the difference image that is the output of the difference pixel extraction unit in the difference image buffer, and the difference pixel extraction unit And a selector 6002 for switching between.

  Note that the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the scope of the invention in the implementation stage. In addition, the embodiments may be appropriately combined as much as possible, and in that case, the combined effect can be obtained. Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if some constituent requirements are deleted from all the constituent requirements shown in the embodiment, the problem described in the column of the problem to be solved by the invention can be solved, and the effect described in the column of the effect of the invention Can be obtained as an invention.

DESCRIPTION OF SYMBOLS 1001 ... Scanner, 1002 ... Layout analysis part, 1003 ... Image component part, 1004 ... Character recognition part, 1005 ... Character separation part, 1006 ... Image dictionary part, 1007 ... Image encoding part, 1008 ... Image file part

JP 2007-86956 A JP 2005-301663 A JP 2007-25815 A

Claims (10)

An analysis means for analyzing the input image;
Based on the analysis result of the analysis means, the first image belonging to the first group is compressed with the first compression parameter, and the second image belonging to the second group is subjected to image degradation based on the first compression parameter. Compression means for compressing with a possible second compression parameter;
An image processing apparatus. The analysis means recognizes each character information included in the input image,
The compression means compresses a first character image belonging to the first group with the first compression parameter based on a character recognition result, and converts a second character image belonging to the second group to the second character image. The image processing apparatus according to claim 1, wherein compression is performed using a compression parameter.   The image processing apparatus according to claim 2, wherein the compression unit compresses a third character image determined to be the same as the second character information by using second encoded information corresponding to the second image. .   The compression means generates first encoded information corresponding to the first character image, compresses the first character image with the first encoded information, and corresponds to the second character image. The second encoded information is generated, the second character image is compressed with the second encoded information, and the third character image is compressed with the second encoded information. The image processing apparatus described.   The compression means generates the first encoded information and the first position information corresponding to the first character image, and the second encoded information and the second corresponding to the second character information. The first character image is compressed with the first encoded information and the first position information, and the second character image is compressed with the second encoded information and the second encoded information. 5. The image processing apparatus according to claim 4, wherein the third character image is compressed using the second position information corresponding to the second encoded information and the third character information.   The image processing apparatus according to claim 2, wherein the analysis unit creates a compressed image for each predetermined character string based on a character recognition result.   The image processing apparatus according to claim 2, wherein the analysis unit compresses a numerical image with the first compression parameter based on a character recognition result.   The image processing apparatus according to claim 2, wherein the analysis unit compresses an image of a number or a sequence of numbers following a specific keyword based on a character recognition result using the first compression parameter. The analysis means analyzes a layout of character / non-character area information included in the input image,
The compression unit compresses an image of a heading, table, or graph belonging to the first group based on a layout analysis result using the first compression parameter, and the heading, table, or graph belonging to the second group. The image processing apparatus according to claim 1, wherein an image not corresponding to the above is compressed with the second compression parameter. Analyze the input image,
There is a possibility that the first image belonging to the first group is compressed with the first compression parameter based on the analysis result, and the second image belonging to the second group is deteriorated by the first compression parameter. An image processing method for compressing with a second compression parameter.

Images