JP2012005076A - Character image compression apparatus and character image restoration apparatus, character image compression method and character image restoration method, and character image compression program and character image restoration program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To compress a character image with a high quality and a low capacitance.SOLUTION: A character recognition section 51 performs character recognition on a character image acquired by a character image acquisition section 50, a linear sum image production section 53 acquires from a dictionary DB 58 a plurality of sample images corresponding to the recognized character and produce a linear sum image with the plurality of sample images. Furthermore, a differential image production section 55 produces a differential image between the character image acquired by the character image acquisition section 50 and the linear sum image, and a storage section 56 stores in storage means 30 the linear sum image and the differential image in association with each other.

Description

  The present invention relates to a character image compression device, a character image restoration device, a character image compression method, a character image restoration method, a character image compression program, and a character image restoration program.

  When storing a large amount of document images, or when transmitting a large amount of form images acquired at a window such as banking operations to a business center or the like, it is preferable that the document images can be compressed with high quality and low capacity.

  In general, there is JPEG (Joint Photographic Experts Group) compression as a method for compressing images. However, when JPEG compression is performed on a document image, block noise may occur around the character pattern and the visibility may be greatly reduced. In particular, when the image quality of the original document image is inferior or the characters are small, the characters are very difficult to see when JPEG compression is performed.

  On the other hand, recently, for example, a compression method using a technique such as Patent Document 1 has been proposed. The technique described in Patent Document 1 separates a document image into a character image and a background image, and performs compression suitable for each. Specifically, the background image is subjected to JPEG compression, while the character image is reduced in character color and converted into a plurality of binary images and subjected to reversible compression. is there.

  Furthermore, a compression method using a character recognition result as in Patent Document 2 has also been proposed. In this technique, characters with high reliability in character recognition results are replaced with character codes and stored. Japanese Patent Laid-Open No. 2004-228561 searches for a character font closest to a character image by searching for a character code as a result of character recognition, sets it as an approximate image, sets it as a predicted image for an input image, obtains a prediction error, and encodes it. Technology is disclosed.

JP 2005-159663 A JP-A-8-295934 JP-A-10-178638

  However, with the technique of the above-mentioned Patent Document 1, visibility cannot always be improved because a part of the character pattern disappears by reducing the color of the character. Further, even if character recognition is used as in the technique of Patent Document 2, the character recognition result itself is not 100% reliable, so that an erroneous image may be restored. Further, even if the technique of Patent Document 3 is used, the document image to be acquired is often deteriorated, so that the difference between the character image of the closest font and the actual image becomes large, and the character image has high accuracy. There is a possibility that (high quality) compression cannot be realized.

  Accordingly, the present invention has been made in view of the above problems, and provides a character image compression device, a character image compression method, and a character image compression program that are capable of compressing a character image with high quality and low capacity. Objective. It is another object of the present invention to provide a character image restoration device, a character image restoration method, and a character image restoration program that can restore a character image that has been compressed with a high quality and a low capacity.

  The character image compression device described in this specification includes a character image acquisition unit that acquires a character image, a character recognition unit that performs character recognition of the character image acquired by the character image acquisition unit, and a character recognition unit. A plurality of sample images corresponding to the recognized character, a linear sum image generation unit that generates a linear sum image using the plurality of sample images, the character image acquired by the character image acquisition unit, A difference image generation unit that generates a difference image with the linear sum image; and a storage unit that associates the information of the linear sum image with the information of the difference image and stores the information in a storage unit.

  The character image restoration device described in the present specification acquires the linear sum image information and the difference image information stored in the storage unit by the storage unit of the character image compression device described in the specification. And a restoration unit that restores the character image from the information acquired by the acquisition unit.

  In the character image compression method described in the present specification, a computer acquires a character image acquisition step of acquiring a character image, a character recognition step of performing character recognition of the character image acquired in the character image acquisition step, and the character A linear sum image generation step of acquiring a plurality of sample images corresponding to the characters recognized in the recognition step and generating a linear sum image using the plurality of sample images, and the character image acquired in the character image acquisition step And a difference image generating step for generating a difference image with the linear sum image, and a storage step for associating the information of the linear sum image with the information of the difference image and storing in the storage means. This is an image compression method.

  The character image restoration method described in the present specification is such that the computer uses the character image compression method described in the specification to store the linear sum image information and the difference image information stored in the storage unit. And a restoration step of restoring the character image from the information acquired in the acquisition step.

  The character image compression program described in the present specification includes a computer, a character image acquisition step of acquiring a character image, a character recognition step of performing character recognition of the character image acquired in the character image acquisition step, and the character A linear sum image generation step of acquiring a plurality of sample images corresponding to the characters recognized in the recognition step and generating a linear sum image using the plurality of sample images, and the character image acquired in the character image acquisition step And a difference image generating step for generating a difference image with the linear sum image, and a storage step for storing the linear sum image information and the difference image information in association with each other and storing them in storage means An image compression program.

  The character image restoration program described in the present specification uses the character image compression program described in the present specification on a computer to store the linear sum image information and the difference image information stored in the storage unit. And a restoration process for restoring the character image from the information acquired in the acquisition process.

  The character image compression apparatus, the character image compression method, and the character image compression program described in this specification have an effect that a character image can be compressed with high quality and low capacity. Further, the character image restoration device, the character image restoration method, and the character image restoration program described in the present specification have an effect that a character image compressed with a high quality and a low capacity can be restored.

It is a figure showing roughly the composition of the information processor concerning one embodiment. It is a figure which shows the hardware constitutions of PC of FIG. A functional block diagram of the PC of FIG. 1 is shown. It is a detailed functional block diagram of a character image compression part and a character image restoration part. It is a figure which shows the data structure of dictionary DB. It is a flowchart which shows the specific processing content of the document image compression process in the information processing apparatus of one Embodiment. It is a flowchart which shows the specific process in step S16 (subroutine of character image compression) of FIG. FIG. 8A to FIG. 8D are diagrams for explaining a method of generating a linear sum image. It is a flowchart which shows the specific process of step S28 of FIG. FIG. 10A and FIG. 10B are diagrams for explaining a method of generating a difference image. FIG. 11A is a diagram schematically showing the data structure of compressed character data, and FIG. 11B is a diagram showing the data structure of character image data. It is a figure which shows the data structure of difference image data. It is a flowchart which shows the restoration process of a document image. It is a flowchart which shows the specific processing content of step S94 of FIG. FIG. 15A to FIG. 15C are diagrams for explaining the comparative example 1. FIG. FIG. 16A to FIG. 16C are diagrams for explaining the comparative example 2. FIG. It is a functional block diagram of PC concerning a modification. FIG. 18A is a diagram illustrating a display example according to the modification, and FIG. 18B is a diagram illustrating a state where the image of FIG. FIG. 19A to FIG. 19C are diagrams showing the calculation procedure of the weighting coefficient in the modified example.

  Hereinafter, an embodiment of a character image compression device and a character image restoration device will be described in detail. FIG. 1 schematically shows the configuration of an information processing apparatus 100 according to an embodiment. As shown in FIG. 1, the information processing apparatus 100 includes a PC 10, a display 12, a keyboard 14, a mouse 16, and a scanner 18.

  The PC 10 has a function of compressing and holding a document image input from the outside, restoring a saved document image in accordance with a user instruction, and displaying the restored document image on the display 12.

  FIG. 2 shows the hardware configuration of the PC 10. As shown in FIG. 2, the PC 10 includes a CPU 90, a ROM 91, a RAM 92, a storage unit (HDD (Hard Disk Drive)) 93, an input / output unit 94, and the like. Has been. In the PC 10, the CPU 90 executes various programs (including a character image compression program and a character image restoration program) stored in the ROM 91 or the HDD 93, thereby realizing the functions of the respective units in FIG. Further, the input / output unit 94 is connected to the display 12, the keyboard 14, the mouse 16, the scanner 18 and the like of FIG.

  Returning to FIG. 1, the display 12 is a display device such as a liquid crystal display, and displays an image input from the PC 10. The keyboard 14 and the mouse 16 are user interfaces used when a user inputs an instruction to the PC 10. The scanner 18 is a device that reads an image (referred to as a document image) of a paper document in which characters, graphics, and the like are described, and inputs them to the PC 10.

  FIG. 3 shows a functional block diagram of the PC 10. As shown in FIG. 3, the PC 10 includes a document image separation unit 20, a background image compression unit 22, a character image compression unit 24 as a character image compression device, a storage unit 30, a background image restoration unit 40, A character image restoration unit 42 as a character image restoration device, a document image generation unit 44, and a display control unit 46 are included.

  The document image separation unit 20 separates the document image read by the scanner 18 into a background image and a character image. Here, the character image is an image obtained by extracting characters included in the document image one by one. The background image is an image of a portion other than the character image.

  The background image compression unit 22 compresses the background image separated by the document image separation unit 20 in the JPEG format or the like, and stores the compressed background data generated by the compression in the storage unit 30. The character image compression unit 24 uses the data stored in the dictionary DB 58 to compress the character image separated by the document image separation unit 20 by a method described later. The character image compressing unit 24 stores the compressed character data generated by the compression in the storage unit 30. Details of the character image compression unit 24 will be described later.

  Here, the storage means 30 and the dictionary DB 58 are realized by the HDD 93 of FIG. The dictionary DB 58 stores a plurality of fonts and images of a plurality of deterioration levels (referred to as sample images (see FIG. 8B)) for a plurality of characters.

  The background image restoration unit 40 reads the compressed background data stored in the storage unit 30 and restores the background image from the compressed background data. The character image restoration unit 42 reads the compressed character data stored in the storage unit 30 and restores the character image based on the compressed character data and the data stored in the dictionary DB 58. Details of the character image restoration unit 42 will be described later.

  The document image generation unit 44 generates (restores) a document image using the background image restored by the background image restoration unit 40 and the character image restored by the character image restoration unit 42. The display control unit 46 displays the document image generated by the document image generation unit 44 on the display 12.

  FIG. 4 shows a detailed functional block diagram of the character image compression unit 24 and the character image restoration unit 42. As shown in FIG. 4, the character image compression unit 24 includes a character image acquisition unit 50, a character recognition unit 51, a color information acquisition unit 52, a linear sum image generation unit 53, a difference image generation unit 55, and a storage. Part 56.

  The character image acquisition unit 50 acquires the character image separated from the document image in the document image separation unit 20 of FIG. The character recognition unit 51 performs character recognition of the character image acquired by the character image acquisition unit 50. The color information acquisition unit 52 acquires color information (here, hue and saturation) from the character image.

  The linear sum image generation unit 53 weights each of the plurality of sample images, and generates a linear sum image from the sum of values obtained by integrating the pixel values of the sample images and the weighting coefficients of the sample images. More specifically, it includes a sample image acquisition unit 54a, a weight coefficient determination unit 54b, and an image generation unit 54c. The sample image acquisition unit 54 a acquires a plurality of sample images corresponding to the characters recognized by the character recognition unit 51 from the dictionary DB 58. In addition, the weighting coefficient determination unit 54b determines the weighting coefficient of each sample image so that the image generated by linear sum using a plurality of sample images is similar to the character image. The image generation unit 54c generates a linear sum image from the sum of values obtained by integrating a plurality of sample images and the weighting coefficient of each sample image.

  The difference image generation unit 55 generates a difference image between the character image acquired by the character image acquisition unit 50 and the linear sum image generated by the linear sum image generation unit 53, and reversibly compresses the difference image to generate the difference image. Generate data.

  The storage unit 56 associates linear sum image information (sample image information and character image data including weighting coefficients used for linear sum image generation) with difference image information (difference image data), and stores the storage unit 30. To store.

  As shown in FIG. 4, the character image restoration unit 42 of FIG. 3 includes an acquisition unit 60, a restoration unit 62, and an output unit 64.

  The acquisition unit 60 acquires linear sum image information and difference image information among the information stored in the storage unit 30. The restoration unit 62 restores a character image from the linear sum image generated from the information of the linear sum image acquired by the acquisition unit 60 and the sample image of the dictionary DB 58 and the difference image generated from the information of the difference image. The output unit 64 outputs the character image restored by the restoration unit 62 to the document image generation unit 44 of FIG.

  Next, the data structure of the dictionary DB 58 will be described with reference to FIG. As shown in FIG. 5, the dictionary DB 58 includes an item “number of character types” and items “character type 1” to “character type p”. In FIG. 5, p is input to the item “number of character types”. Here, the character type means the character itself (for example, “A”, “I”, etc.), and includes one character type including fonts and images with different degrees of deterioration. The character type is generally called a character category, and the number of character types is usually about 4000 for Japanese.

  In addition, as shown in FIG. 5, the dictionary DB 58 includes an item “character code” and an item “number of character images” for each character type. In addition, an item “ID”, an item “character image (data)”, and an item “character feature (vector)” are provided corresponding to a plurality of character images. In the “character image” item, for example, images (also referred to as “sample images”) having different degrees of deterioration of fonts and images as shown in FIG. 8B are registered, and in the “ID” item, IDs such as A1 to D3 shown in FIG. 8B are registered.

  As the character image, a normalized image, for example, a gray image is registered. The number of dimensions of the character feature vector is also constant regardless of the character image. Therefore, the information for each character image is an area having a fixed size. The character feature vector may be obtained from a binary image instead of a gray image. Further, the degree of image degradation can be obtained, for example, by acting on a character image while changing the variance of the Gaussian filter in multiple stages.

  Next, details of the document image compression process and the document image restoration process in the information processing apparatus 100 according to the present embodiment will be described with reference to FIGS.

  FIG. 6 is a flowchart showing specific processing contents of the document image compression processing in the information processing apparatus 100 according to the present embodiment. In the process of FIG. 6, first, in step S <b> 10, the document image separation unit 20 acquires a document image read by the scanner 18. Next, in step S12, the document image separation unit 20 separates the character image and the background image. As this separation method, for example, a method described in JP-A-2005-159663 can be employed.

  Next, in step S14, the background image compression unit 22 executes a process of compressing the background image. In this case, the background image compression unit 22 compresses the background image, for example, in the JPEG format, and generates compressed background data.

  In step S16, the character image compression unit 24 executes a character image compression subroutine. In this character image compression subroutine, the processing of FIG. 7 is executed.

  FIG. 7 is a flowchart showing specific processing in step S16 (a subroutine for character image compression). In the process of FIG. 7, first, in step S <b> 20, the character image acquisition unit 50 of the character image compression unit 24 acquires all the character images separated from the document image by the document image separation unit 20. The character image acquired here is assumed to be a color character image.

  Next, in step S22, the character recognition unit 51 performs character recognition processing using the character image (color character image) acquired in step S20. The character recognition result obtained as a result is the number of recognition result characters and the character information of each recognition result. The character information of the recognition result includes a character code and character coordinates.

  Next, in step S23, the color information acquisition unit 52 acquires an unprocessed character image. Next, in step S24, the color information acquisition unit 52 acquires color information of the acquired character image. Here, the color information of the character image is a value of hue and saturation. In this case, first, the color information acquisition unit 52 generates a lightness image (F) from a color original image (I). Then, the color information acquisition unit 52 performs hue and saturation for pixels that have a brightness value greater than a predetermined threshold among the pixels of the character image, that is, for pixels that are displayed more than a predetermined black when the character image is a gray image. Calculate and sample the degree value. In addition, the color information acquisition unit 52 obtains the hue and saturation values having the highest appearance frequency as a result of the sampling, and uses them as the color information of the character image. The color and saturation acquisition method is not limited to the above method, and the color information acquisition unit 52 obtains the average value of the color and saturation of the pixels of the character image, and acquires this as the color information of the character image. It is also good.

  Next, in step S26, the linear sum image generation unit 53 generates a linear sum image. Hereinafter, a method for generating a linear sum image will be described in detail with reference to FIGS. 8 (a) to 8 (d).

FIG. 8A shows a character image of the character “proof” as an example of a character image (lightness image) F for one character. As a result of the character recognition (step S22) of the character image F, when it is correctly recognized as “certificate”, the sample image acquisition unit 54a obtains a plurality of sample images corresponding to the characters “certification” from the dictionary DB 58, Obtained as shown in FIG. Then, the weighting coefficient determining section 54b is directed to the character image F, and calculates the distance between the sample images, the distance the higher close in the N sample images f _1, f _2, ..., determining f _N. Here, the distance between the character image F and the sample image can be calculated using an identification function value using the character feature vector of the sample image. In FIG. 8C, an example of N = 3 is shown.

Next, the weighting factor determination unit 54b obtains the optimal similar image of the linear sum image for the character image F. Specifically, the top N sample images f _1, f _2, ..., the weighting factor alpha ₁ for f _N, alpha _2, ..., the alpha _N, obtained from the following equation (1).

Here, if the problem of the above equation (1) is regarded as a constrained least square problem, the weight coefficients α ₁ , α ₂ ,..., Α _N can be calculated as the solution. Specifically, the weighting factor determination unit 54b calculates α ₁ , α ₂ ,..., Α _N based on the following equation (2).

なお、上式（２）では、Ｃ＝｛（Ｆ−ｆ_k）・（Ｆ−ｆ_ｌ）｝であるものとする。

In the above formula (2), C = {(F−f _k ) · (F−f ₁ )}.

When the weighting coefficients α ₁ , α ₂ ,..., Α _N are determined as described above, the image generation unit 54c can generate a linear sum image. As an example, as shown in FIG. 8C, if the sample image ID is A3, the weighting factor is 0.60, the weighting factor of A2 is 0.31, and the weighting factor of C3 is 0.09. The image generation unit 54c can generate a linear sum image as shown in FIG. 8D from the sum of values obtained by integrating the pixel values of the sample images and the weighting coefficients of the sample images. After the linear sum image is generated in this way, the process proceeds to step S28 in FIG.

  In step S28 of FIG. 7, the difference image generation unit 55 executes a subroutine for generating a difference image. In this subroutine, the difference image generation unit 55 executes a process as shown in FIG. Note that the processing in step S28 can be simply described by subtracting the linear sum image generated in step S26 from the character image F as shown in FIG. 10A to generate a difference image as shown in FIG. It is processing to do.

  In the process of FIG. 9, first, in step S40, the difference image generation unit 55 generates a positive difference image and a negative difference image. The positive difference image is an image showing pixels that need to be made blacker in order to bring the linear sum image closer to the original character image, and the negative difference image is the original character image. It is an image showing a pixel that needs to be whiter in order to be closer to.

  Specifically, the difference image generation unit 55 obtains the difference image D from the following equation (3).

Next, the difference image generation unit 55 decomposes D into a positive difference image D ₊ and a negative difference image D ₋ . Specifically, the difference image generation unit 55 obtains D ₊ and D ₋ from the following expressions (4) and (5). Note that m represents a pixel number in the row direction, and n represents a pixel number in the column direction.

Next, in step S42, it is determined whether or not the pixel values (total pixel values) of the entire positive difference image are equal to or less than a predetermined threshold th _α . Here, as the threshold th _α , for example, the maximum value in the range of pixel values of the difference image that can be taken when the linear sum image is correctly generated (obtained by a prior experiment or simulation) is adopted. To do. If the determination is negative, that is, if it is determined that there is a high possibility that an appropriate linear sum image is not generated in step S26 because the positive difference image is too large, the process proceeds to step S78. To do. In step S78, the positive difference image is set to the character image F, the negative difference image is set to 0 (Null), and the weighting coefficient is cleared. In this way, when step S78 is performed, the character image itself is handled as a positive difference image without performing image compression using the linear sum image generated in step S26 described above. After step S78 is performed, all the processes in FIG. 9 are terminated, and the process proceeds to step S30 in FIG.

On the other hand, if the determination in step S42 is affirmative, the process proceeds to step S44. In step S44, the difference image generation unit 55 sets the row number m (0 ≦ m ≦ y _max ) of the pixels of the difference image to 0 (m = 0). Next, in step S46, the difference image generation unit 55 determines whether or not the line number m is equal to or less than the maximum value y _{max of the} line numbers. If the determination here is affirmed, the process proceeds to step S48. If the determination is negative, the process proceeds to step S60. When the process proceeds to step S48, the difference image generation unit 55 sets the column number n (0 ≦ n ≦ x _max ) of the pixels of the difference image to 0 (n = 0). Next, in step S50, the difference image generation unit 55 determines whether or not the column number n is equal to or less than the maximum value x _{max of the} column numbers. If the determination is affirmed, the process proceeds to step S52. If the determination is negative, m is incremented by 1 (m ← m + 1) in step S58, and the process returns to step S46.

When the process proceeds to step S52, the difference image generation unit 55 determines whether the pixel value of m rows and n columns is equal to or greater than a predetermined threshold th _β . Here, as the threshold value th _β , a pixel value can be adopted such that even if the pixel value is approximated to 0, there is almost no difference in appearance between the restored image and the image before compression. When the determination here is negative, that is, when the pixel value of m rows and n columns is smaller than the threshold th _β and can be regarded as almost 0, in step S54, the difference image generation unit 55 determines that m rows. The pixel values of the n columns are set to 0, and the process proceeds to step S56. In this way, by setting the pixel value to 0 when the pixel value of m rows and n columns can be regarded as almost 0, the data amount of the difference image can be reduced.

  On the other hand, if the determination in step S52 is affirmed, the difference image generation unit 55 proceeds to step S56 without passing through step S54, that is, while maintaining the pixel value of m rows and n columns.

In step S56, the difference image generation unit 55 increments the column number n by 1, and then returns to step S50. Then, the difference image generation unit 55, by repeating the steps S46 to S56, steps from the pixels of row zero and column zero executes the process up to the pixels of y _max rows x _max columns, the determination in step S46 is negative Thus, the process proceeds to step S60.

Note that the subsequent processing from step S60 to step S76 is processing for the negative difference image, but is performed in the same manner as the processing for the positive difference image from step S42 to step S58 described above. In step S60, the absolute value of the pixel value (total sum of pixel values) of the entire negative difference image is compared with the threshold th _α, and in step S70, the absolute value of the pixel value in m rows and n columns and the threshold th _β are compared. Is different from the processing for the positive difference image described above. If the determination in step S60 is negative, the process in step S78 is completed, and all the processes in FIG. 9 are terminated. If the determination in step S64 is negative, the process in FIG. End all processing.

  When all the processes in FIG. 9 are completed, the process proceeds to step S29 in FIG. In step S29, the difference image generation unit 55 reversibly compresses the difference image generated in step S28 to generate difference image information (difference image data). Next, in step S30, the color information acquisition unit 52 determines whether acquisition of all unprocessed characters has been completed. If the determination is negative, the process returns to step S23, and the processes and determinations of steps S23 to S30 are repeated until the determination of step S30 is affirmed. On the other hand, if the determination in step S30 is affirmative, the entire process in FIG. 7 is terminated, and the process proceeds to step S18 in FIG. Note that each piece of data generated at the stage when the processing of FIG. 7 is completed is referred to as “compressed character data”. That is, as schematically shown in FIG. 11A, the compressed character data includes character image data including information on a sample image, weight coefficient, color information, and the like used for linear sum image generation, and difference image data. Contains.

  Here, the character image data and difference image data shown in FIG.

  As shown in FIG. 11B, the character image data has an item “number of characters” at the head, and thereafter, “code information for one character” is arranged by the number of characters. In the item “code information for one character”, “character coordinates” (for example, the coordinates of the upper left point and the lower right point of the character), “character color information”, and “character code” are described. Is done. In addition, the number of weighting factors used to generate the linear sum image is described, and the ID of the sample image corresponding to the weighting factor and its weighting factor are described. Note that the character color information is composed of two numerical values, hue and saturation.

  On the other hand, as shown in FIG. 12, the difference image data has an item “number of characters” at the head, and thereafter, information of “difference image for one character” is arranged by the number of characters. The “difference image for one character” describes the data size and character coordinates (for example, the coordinates of the upper left point and the lower right point of the character). Then, the size of the positive difference image data is written, and then the positive difference image data is arranged. Further, the size of the negative difference image data is described, and thereafter the negative difference image data is arranged. The difference image data is data compressed by lossless compression.

  As can be seen from FIG. 11 and FIG. 12, the coordinates are described in “Code information for one character” in FIG. 11 and “Difference image for one character” in FIG. The code information and the difference image are associated with each other.

  Returning to FIG. 6, in step S <b> 18, the background image compression unit 22 stores the compressed background data generated in step S <b> 14 in the storage unit 30 and the storage unit 56 (see FIG. 4) of the character image compression unit 24. The compressed character data is stored in the storage means 30.

  Next, a document image restoration process using the compressed background data and the compressed character data stored in the storage unit 30 as described above will be described with reference to the flowchart of FIG.

  In the process of FIG. 13, first, in step S90, the acquisition unit 60 determines whether or not there is an instruction to restore the document image from the user. While the determination is negative, step S90 is repeated, and the process proceeds to step S91 when the determination in step S90 is affirmed.

  In step S <b> 91, the acquisition unit 60 acquires the compressed background data corresponding to the document image designated in the restoration instruction by the user from the storage unit 30. Next, in step S <b> 92, the acquisition unit 60 acquires the compressed character data corresponding to the document image specified in the restoration instruction by the user from the storage unit 30.

  Next, in step S93, the restoration unit 62 restores the background image using the compressed background data, that is, expands the image on the memory (here, referred to as the memory A for convenience). Next, in step S94, the restoration unit 62 executes a subroutine for restoring a character image.

  Specifically, the restoration unit 62 executes processing according to the flowchart of FIG. In FIG. 14, first, in step S102, the restoration unit 62 acquires the compressed character data shown in FIG.

  Next, in step S104, the restoration unit 62 expands the linear sum image on a memory B different from the memory A, and overwrites the memory A according to the coordinate information described in the compressed character data of FIG. More specifically, in step S104, the restoration unit 62 first expands the linear sum image in the memory B, and creates a linear sum image for the number of characters written in the compressed character data. At this time, based on the ID written in the compressed character data, the restoration unit 62 searches the dictionary DB 58 to obtain a corresponding sample image, adds a weighting factor to the sample image, and sequentially adds the sample images to linearly. Create a Japanese image.

  Next, in step S106, the restoration unit 62 develops the difference image on the memory B. Here, since the difference image data is losslessly compressed, the restoration unit 62 decodes and decompresses the difference image data. In this case, the restoration unit 62 performs an operation of adding the pixel value of the difference image on the memory B using the decoded difference image information. The positive difference image is added and the negative difference image is subtracted. It should be noted that a portion where the value of the linear sum image is smaller than a predetermined threshold is made transparent.

  Next, in step S108, the restoration unit 62 colorizes the character image generated in step S108. In this colorization, the restoration unit 62 colorizes the character image using the hue and saturation values obtained from the color information in the character image data as they are.

  As described above, when the character image is restored from the compressed character data acquired in step S102, the entire process in FIG. 14 is terminated, and the process proceeds to step S96 in FIG. At this stage, the document image in which the background image and the character image are combined is restored.

  In step S96, the output document 64 outputs the restored document image to the display control unit 46 in FIG. When the processing of FIG. 13 is completed as described above, the display control unit 46 displays the document image on the display 12, so that the user can view and use the document image.

  As described above in detail, according to the present embodiment, the character recognition unit 51 performs character recognition of the character image acquired by the character image acquisition unit 50, and the linear sum image generation unit 53 corresponds to the recognized character. A plurality of sample images to be obtained are acquired from the dictionary DB 58, and a linear sum image using the plurality of sample images is generated. Further, the difference image generation unit 55 generates a difference image between the character image acquired by the character image acquisition unit 50 and the linear sum image, and the storage unit 56 associates the linear sum image and the difference image, and stores them. To store. Thus, in this embodiment, since a linear sum image is generated using a plurality of sample images, a linear sum image that approximates the character image can be generated even if the character image changes in various ways. it can. Thereby, a character image can be compressed with high precision (high quality). In the present embodiment, since a linear sum image approximate to a character image can be generated, the capacity of the difference image can be reduced. As a result, the character image can be compressed to a low capacity. Furthermore, even if the character recognition accuracy is not 100%, it is possible to supplement the error of character recognition by using the difference image as in the present embodiment. From this point of view, high accuracy (high quality) ) Character images can be compressed.

  Here, a comparative example will be described. 15A to 15C show an example (Comparative Example 1) in which a character image is JPEG compressed. Among these, as shown in the left figure of FIG. 15A, in the image subjected to JPEG compression, block noise occurs around the character pattern as shown in the right figure (enlarged figure) of FIG. 15A. As a result, the visibility may be significantly reduced. In this case, the occurrence of block noise as shown in FIG. 15B causes a shift as compared with the original image shown in FIG.

  FIGS. 16A to 16C show an example (comparative example 2) in which the character color of the character image is reduced and reversible compression is performed on a plurality of binary images. Among these, as shown in the left diagram of FIG. 16A, in the image subjected to the color reduction process, a part of the character pattern disappears as shown in the right diagram (enlarged diagram) of FIG. As a result, the visibility may be significantly reduced. In this case, as shown in FIG. 16B, a part of the character disappears, so that a deviation from the original image shown in FIG. 16C occurs.

  On the other hand, in this embodiment, even if the original image is compressed, almost the same image can be restored, so that the visibility of characters can be maintained equivalent to that of the original image.

  In the present embodiment, the linear sum image generation unit 53 weights each of the plurality of sample images corresponding to the characters recognized by the character recognition unit 51, and the pixel value of each sample image and each sample image A linear sum image is generated from the sum of values obtained by integrating the weighting factors. Thus, in this embodiment, by determining the weighting factor for each sample image, a linear sum image that is similar (approximate) to the character image can be generated.

  Further, in the present embodiment, the weighting coefficient determination unit 54b uses each of the restricted least squares problems as a solution of the constrained least square problem so that the distance between the feature vector of the character image and the feature vector of the generated image by linear sum is the minimum distance. Since the weight coefficient of the sample image is determined, a linear sum image that is similar (approximate) to the character image with high accuracy can be generated by performing an appropriate calculation.

  In the present embodiment, a plurality of sample images having different font types and at least one of deterioration levels are prepared for the same character. Therefore, the character image to be compressed may be various fonts, A linear sum image that is similar (approximate) to a character image to be compressed can be generated even if the degree of image degradation varies.

  In the present embodiment, the acquisition unit 60 of the character image restoration unit 42 acquires linear sum image information (character image data) and difference image information (difference image data) stored in the storage unit 70. Then, the restoration unit 62 restores the character image from the information. Thus, by performing restoration using a linear sum image similar to a character image with high accuracy and a difference image, the original character image can be restored with high accuracy.

  In the above-described embodiment, the case where the document image is compressed and then restored so as to be closer to the original document image is described. However, the present invention is not limited to this. If the character image portion of the document image is unclear, the character image portion may be clearly displayed in response to an instruction from the user (or automatically). A specific example (modified example) of such a method will be described below.

  FIG. 17 is a functional block diagram of the PC 10 according to the modification. As can be seen by comparing FIG. 17 with FIG. 3, FIG. 17 is different from the above embodiment in that a definition input receiving unit 96 as a receiving unit is provided. The definition input receiving unit 96 has a function of receiving the input of the definition and outputting it to the character image restoration unit 42 when the user inputs the definition via the mouse 16 or the keyboard 14.

  FIG. 18A shows a screen display example of the display 12 in this modification. As shown in FIG. 18A, in this modification, a slider control 98 for sharpness control is displayed on the lower side of the screen. In the slider control 98, the user performs a sharpness input by sliding the knob 99 left and right using the mouse 16 or the like.

  As a premise of this modification, it is assumed that the ID of the sample image registered in the dictionary DB 58 includes information on the degree of deterioration. Specifically, if the ID is represented by 32 bits, the first 4 bits (values range from 0 to 15) are assigned to the degree of deterioration. The degree of deterioration is defined such that 0 is “no deterioration”, and the deterioration increases as the number increases. A character code is assigned to the next 16 bits of the ID, and a serial number for the same character code is assigned to the remaining 12 bits. For example, if the character is “go” (character code: 0x8CE4), the deterioration degree is 1, and the serial number is 7, the ID is represented as “18CE4007” in hexadecimal notation. The degree of deterioration is a value based on the variance of the Gaussian filter that acts on the character image, as in the above embodiment.

  Under the above premise, the character image restoration unit 42 starts restoration of the linear sum image when receiving an input of the sharpness (the sharpness is assumed to take a value of 0 (low) to 1 (high)). Here, it is assumed that there are four sample images (weighting factors) used for restoration of the linear sum image, and weighting factors as shown in FIG. 19A are set for these IDs.

  When restoring the linear sum image, the character image restoration unit 42 adds (1−sharpness) to the weight coefficient of the image whose degradation degree is not zero. FIG. 19B shows the calculation result when the sharpness is 0.8.

  Next, the character image restoration unit 42 normalizes each weight coefficient so that the sum of the four weight coefficients is 1. The normalized weighting coefficient is shown in FIG. 19 (c). Then, the character image restoration unit 42 generates a linear sum image using the normalized weighting coefficient shown in FIG. The other processes are the same as in the above embodiment.

  By performing processing in consideration of such sharpness, it is possible to generate a linear sum image with priority given to an image with a degradation degree of 0. Therefore, an unclear image as shown in FIG. As shown in FIG. As a result, an image clearer than the original image can be displayed on the display 12.

  In the above modification, a case has been described in which (1-definition) is uniformly integrated with respect to a weighting factor of an image whose degradation degree is not 0. However, the present invention is not limited to this. For example, in consideration of the degree of deterioration (1 to 15), an operation may be performed on the weighting coefficient so that the weighting coefficient decreases as the degree of deterioration increases.

In the above embodiment, when the difference image is generated, if the absolute value of the sum of the pixel values of the difference image is larger than the predetermined threshold th _α in steps S42 and S60, the compression using the sample image is performed. Explained the case of not being. However, the present invention is not limited to this, and the compression using the sample image may always be performed regardless of the total sum of the pixel values of the difference image.

In the above embodiment, when the absolute value of the pixel value in m rows and n columns is smaller than the predetermined threshold th _β in steps S52 and S70, the pixel value is set to 0, thereby reducing the data amount of the difference image. I decided to make it smaller. However, the present invention is not limited to this, and the actual pixel value may be maintained regardless of whether the pixel value of each pixel is close to 0 or not.

  In the above embodiment, the case where the character image is a color has been described. However, the present invention is not limited to this, and the document image read by the scanner 18 may be a monochrome image. When the document image is always a monochrome image, the color information acquisition unit 52 in FIG. 4 may be omitted and the processing of the color information acquisition unit 52 may be omitted.

  In the above-described embodiment, the case where the document image is acquired using the scanner 18 has been described. However, the present invention is not limited to this, and the document image may be acquired via a network such as the Internet. Further, instead of the scanner, a document image may be acquired from a digital camera or a multifunction device having a scanner function.

  In the above embodiment, the case where the compression and restoration of the document image is performed by one PC has been described. However, the present invention is not limited to this, and the document image is compressed by one PC and the document image is compressed by another PC. Restoration may be performed. In this case, one PC and another PC may be connected via a network such as the Internet or a LAN (local area network), and compressed background data and compressed character data may be exchanged via the network. . Alternatively, a portable storage medium such as a USB memory may be used for data exchange between one PC and another PC. In this case, one PC and another PC may be connected via a network or may not be connected.

  In the above-described embodiment, the character image compression program and the character image restoration program of the present case are incorporated in the PC 10 to realize each function of each unit in FIG. However, the present invention is not limited to this. For example, a server computer connected to a communication network such as the Internet is used as the character image compression device and the character image restoration device of the present case. A service for storing data may be provided from a server computer (ASP (Application Service Provider)).

  The above processing functions can be realized by a computer. In that case, a program describing the processing contents of the functions that the processing apparatus should have is provided. By executing the program on a computer, the above processing functions are realized on the computer. The program describing the processing contents can be recorded on a computer-readable recording medium.

  When the program is distributed, for example, it is sold in the form of a portable recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory) on which the program is recorded. It is also possible to store the program in a storage device of a server computer and transfer the program from the server computer to another computer via a network.

  The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing according to the program. Further, each time the program is transferred from the server computer, the computer can sequentially execute processing according to the received program.

  The above-described embodiment is an example of a preferred embodiment of the present invention. However, the present invention is not limited to this, and various modifications can be made without departing from the scope of the present invention.

In addition, the following additional notes are disclosed regarding the above description.
(Supplementary Note 1) Corresponding to a character image acquisition unit that acquires a character image, a character recognition unit that performs character recognition of the character image acquired by the character image acquisition unit, and a character recognized by the character recognition unit A linear sum image generation unit that acquires a plurality of sample images and generates a linear sum image using the plurality of sample images, a difference image between the character image acquired by the character image acquisition unit and the linear sum image A character image compression apparatus comprising: a difference image generation unit that generates the image; and a storage unit that associates the information of the linear sum image and the information of the difference image and stores the information in a storage unit.
(Supplementary Note 2) The linear sum image generation unit weights each of the plurality of sample images, and generates a linear sum image from a sum of values obtained by integrating pixel values of the sample images and weighting coefficients of the sample images. The character image compression apparatus according to Supplementary Note 1, wherein:
(Supplementary note 3) The character image compression device according to supplementary note 1 or 2, wherein the plurality of sample images are images in which at least one of a font type and a deterioration degree is different with respect to the same character.
(Additional remark 4) The acquisition part which acquires the information of the said linear sum image and the information of the said difference image which the said storage part of the character image compression apparatus in any one of Additional remarks 1-3 stored in the said storage means; A character image restoration apparatus comprising: a restoration unit that restores the character image from information acquired by the acquisition unit.
(Additional remark 5) It further has a reception part which receives the input of the definition of the character image which the restoration part restores, and the restoration part uses a weight coefficient of a sample image contained in information of the linear sum image as the reception part. The character image restoring device according to appendix 4, wherein the character image is changed based on the sharpness received by the user.
(Additional remark 6) The said restoration part considers the deterioration degree of the said sample image, when changing the said weighting coefficient, The character image restoration apparatus of Additional remark 5 characterized by the above-mentioned.
(Additional remark 7) The character image acquisition process in which a computer acquires a character image, the character recognition process of performing the character recognition of the said character image acquired at the said character image acquisition process, and the character recognized by the said character recognition process A linear sum image generation step of acquiring a plurality of corresponding sample images and generating a linear sum image using the plurality of sample images, the character image acquired in the character image acquisition step, and the linear sum image A character image compression method that executes a difference image generation step of generating a difference image, a storage step of associating the information of the linear sum image and the information of the difference image and storing the information in a storage unit.
(Supplementary Note 8) In the linear sum image generation step, each of the plurality of sample images is weighted, and a linear sum image is generated from a sum of values obtained by integrating the pixel values of the sample images and the weighting coefficients of the sample images. 8. The character image compression method according to appendix 7, wherein
(Supplementary note 9) The character image compression according to any one of Supplementary notes 7 and 8, wherein the plurality of sample images are images in which at least one of a font type and a deterioration degree is different with respect to the same character. Method.
(Additional remark 10) A computer acquires the information of the said linear sum image stored in the said storage means, and the information of the said difference image using the character image compression method in any one of additional marks 7-9 A character image restoration method that executes an acquisition step and a restoration step of restoring the character image from the information acquired in the acquisition step.
(Additional remark 11) The acceptance process which receives the input of the definition of the character image decompress | restored by the said reconstruction process is further included, and the said restoration process WHEREIN: The weighting coefficient of the sample image contained in the information of the said linear sum image is said reception The character image restoration method according to appendix 10, wherein the character image is changed based on the sharpness received in the process.
(Additional remark 12) In the said restoration process, when changing the said weighting coefficient, the degradation degree of the said sample image is considered, The character image restoration method of Additional remark 11 characterized by the above-mentioned.
(Additional remark 13) The character image acquisition process which acquires a character image to a computer, the character recognition process which performs the character recognition of the said character image acquired at the said character image acquisition process, and the character recognized by the said character recognition process A linear sum image generation step of acquiring a plurality of corresponding sample images and generating a linear sum image using the plurality of sample images, the character image acquired in the character image acquisition step, and the linear sum image A character image compression program for executing a difference image generation step of generating a difference image, a storage step of associating the information of the linear sum image and the information of the difference image and storing the information in a storage unit.
(Supplementary Note 14) In the linear sum image generation step, each of the plurality of sample images is weighted, and a linear sum image is generated from a sum of values obtained by integrating the pixel values of the sample images and the weight coefficients of the sample images. 14. The character image compression program according to appendix 13, wherein
(Supplementary note 15) The character image compression program according to supplementary note 13 or 14, wherein the plurality of sample images are images in which at least one of a font type and a deterioration degree is different with respect to the same character.
(Additional remark 16) The information of the said linear sum image and the information of the said difference image which were stored in the said storage means are acquired to a computer using the character image compression program in any one of Additional remarks 13-15 A character image restoration program that executes an acquisition step and a restoration step of restoring the character image from the information acquired in the acquisition step.
(Additional remark 17) The acceptance process which receives the input of the definition of the character image restored at the restoration process further includes the weight coefficient of the sample image contained in the information of the linear sum image in the restoration process. The character image restoration program according to appendix 16, wherein the character image restoration program is changed based on the sharpness received in the process.
(Supplementary note 18) The character image restoration program according to supplementary note 17, wherein in the restoration step, the deterioration degree of the sample image is taken into account when the weighting coefficient is changed.

24 character image compression unit (character image compression device)
30 Storage means 42 Character image restoration unit (character image restoration device)
50 character image acquisition unit 51 character recognition unit 53 linear sum image generation unit 54c image generation unit 55 difference image generation unit 56 storage unit 60 acquisition unit 62 restoration unit

Claims (10)

A character image acquisition unit for acquiring a character image;
A character recognition unit that performs character recognition of the character image acquired by the character image acquisition unit;
A linear sum image generation unit that acquires a plurality of sample images corresponding to characters recognized by the character recognition unit and generates a linear sum image using the plurality of sample images;
A difference image generation unit that generates a difference image between the character image acquired by the character image acquisition unit and the linear sum image;
A character image compression apparatus comprising: a storage unit that stores the linear sum image information and the difference image information in association with each other and stores the information in a storage unit. The linear sum image generation unit includes:
2. The character according to claim 1, wherein each of the plurality of sample images is weighted, and a linear sum image is generated from a sum of a value obtained by integrating a pixel value of the sample image and a weight coefficient of each sample image. Image compression device.   The character image compression apparatus according to claim 1, wherein the plurality of sample images are images in which at least one of a font type and a degree of deterioration is different with respect to the same character. An acquisition unit that acquires information on the linear sum image and information on the difference image stored in the storage unit by the storage unit of the character image compression device according to any one of claims 1 to 3.
A character image restoration apparatus comprising: a restoration unit that restores the character image from information acquired by the acquisition unit. A reception unit that receives an input of the definition of the character image restored by the restoration unit;
The character image restoration device according to claim 4, wherein the restoration unit changes a weight coefficient of a sample image included in the information of the linear sum image based on the sharpness received by the reception unit. .   The character image restoration device according to claim 5, wherein the restoration unit considers the degree of deterioration of the sample image when changing the weighting factor. Computer
A character image acquisition step of acquiring a character image;
A character recognition step of performing character recognition of the character image acquired in the character image acquisition step;
A linear sum image generation step of acquiring a plurality of sample images corresponding to the characters recognized in the character recognition step, and generating a linear sum image using the plurality of sample images;
A difference image generation step of generating a difference image between the character image acquired in the character image acquisition step and the linear sum image;
A character image compression method that executes a storing step of associating the information of the linear sum image with the information of the difference image and storing the information in a storage unit. Computer
An acquisition step of acquiring information on the linear sum image and information on the difference image stored in the storage unit using the character image compression method according to claim 7;
A character image restoration method that performs a restoration step of restoring the character image from the information acquired in the acquisition step. On the computer,
A character image acquisition step of acquiring a character image;
A character recognition step of performing character recognition of the character image acquired in the character image acquisition step;
A linear sum image generation step of acquiring a plurality of sample images corresponding to the characters recognized in the character recognition step, and generating a linear sum image using the plurality of sample images;
A difference image generation step of generating a difference image between the character image acquired in the character image acquisition step and the linear sum image;
A character image compression program for executing a storage step of storing the linear sum image information and the difference image information in association with each other in a storage unit. On the computer,
Using the character image compression program according to claim 9, an acquisition step of acquiring information on the linear sum image and information on the difference image stored in the storage unit;
A character image restoration program that executes a restoration step of restoring the character image from the information obtained in the obtaining step.

Images