JP6743092B2 - Image processing apparatus, image processing control method, and program - Google Patents

Description

The present invention relates to an image processing device, an image processing device control method, and a program.

In recent years, with the spread of color printers and color scanners, the number of colorized documents has increased, and there is an opportunity to scan these documents and save them as electronic files or send them to third parties via the Internet. It is increasing. However, since full-color data remains a heavy load on a storage device and a line, it is necessary to perform compression processing to reduce the amount of data.

Conventionally, as a method of compressing a color image, for example, a method of compressing a binary image having a pseudo gradation by error diffusion or the like, a method of compressing in a JPEG format, or a ZIP compression by converting into an 8-bit palette color. And LZW compression method.

Further, in Patent Document 1, a character region included in an input image is detected, the detected character portion is converted into a binary image, and MMR compression (binary lossy compression) is performed, and a file is provided together with character color information of each character. Save to. Further, the character portion on the input image is filled with the surrounding color is used as the background image, the resolution is reduced, and the image is JPEG-compressed (irreversibly compressed) and stored in the file. The file compressed by this compression method can obtain high quality in the character area and can also have a high compression rate.

JP, 2002-077633, A

In Patent Document 1, in a binary image obtained by binarizing an input image, the size (width or height) of a collection of black pixels, and whether there are collections of black pixels of similar size in the vicinity, etc. Based on the above, it is determined whether the collection of each black pixel seems to be a character, and the character area is detected.

On the other hand, when a method of performing area determination based on a binary image as in Patent Document 1 is applied to an input image in which it is difficult to separate a character and a background by simple binarization, a character is formed. Pixel identification becomes difficult. For example, when performing simple binarization on a black character (a character image having a large density difference between the character and the background) on a white background, it is easy to separate the background pixel and the character pixel. On the other hand, when binarizing a black character (a character image image in which the density difference between the character and the background is small) on a dark background, it is difficult to separate the background pixel and the character pixel. In particular, when a character having a dark background is binarized with a threshold value smaller than the background density, the binary character image is blackened. At this time, if the size of the background area of high density is about the same as the size of the character, the binary image in the state where the background and the character are crushed in black and binarized is erroneously determined as a character pixel portion. In some cases it may end up. For example, when a document in which a part of a character string is marked with a dark marker pen is scanned and the scanned image is binarized, the entire portion marked with the marker pen may be black. If the size of the part marked with the marker pen is close to the character size, the entire pixels of the part marked with the marker pen are binarized and treated as one character. In other words, all the black pixels in the black-crushed area may be treated as character pixels during binarization.

In order to solve the above-mentioned problems, the image processing apparatus of the present invention includes a first determination unit that determines whether an attribute of an area in an image is at least a character attribute or an image attribute, and the character attribute. And a second determination unit that determines whether the image of the area determined by the first determination unit is a character image that is easily separated from the background or a character image that is difficult to be separated from the background, and the character attribute If the image of the area determined to be present is a character image that can be easily separated from the background, the second determination means determines that the image of the area has the character attribute by performing a binary compression process. If the second determination means determines that the image of the region is a character image that can be easily separated from the background , the image processing device performs a multi-value compression process on the image of the region, and determines that the image has the character attribute. And a generating unit that generates a file of the image based on at least the image of the area that has been determined by the first determining unit and on which one of the binary compression process and the multi-value compression process has been executed. Characterize.

According to the embodiment of the present invention, it is possible to change whether to perform the binary compression processing or the multi-value compression processing depending on the type of the area determined to have the character attribute.

Block diagram showing image processing system Hardware configuration of MFP in Embodiment 1 Block diagram of the area determination unit 2 in the first embodiment FIG. 3 is a diagram for explaining area determination in the first embodiment. Block diagram of image compression processor Block diagram of image decompression processor Input image sample and output image sample Flowchart of area determination in Example 1 Example of input image in Example 1 Example of input image in Example 2 Block diagram of an edge detector in the fourth embodiment Edge extraction sample diagram Flowchart of edge extraction in the fourth embodiment Block diagram of an edge detection unit in the fifth embodiment 1 2 is a block diagram of an edge detection unit according to the fifth embodiment. Sample diagram in Example 1

(Example 1)
FIG. 1 is a schematic diagram illustrating a system configuration according to the first embodiment. In FIG. 1, a multi-function peripheral (MFP) 101 and a computer (hereinafter, PC) 102 are connected via a network 103.

Dotted lines 104 and 105 show the flow of processing, and 104 shows the processing in which the user uses the scanner of the MFP 101 to read a paper document. At that time, the user can use the user interface (203 in FIG. 2) of the MFP 101, which will be described later, to set the destination (for example, the PC 102) to which the scan image is to be transmitted and various settings related to the scan and the transmission. As the various settings, the user can specify the resolution, compression ratio, data format (for example, JPEG, TIFF, PDF, PDF high compression, PDF high compression (with OCR result)) and the like. In the present embodiment, a case will be described where PDF high compression (with OCR result) is designated as the data format. The technical details of PDF high compression will be described later. Reference numeral 105 denotes a process for generating data using the software or hardware function of the MFP 101 based on various designated settings and transmitting the data to the designated destination. Here, since the image transmitted to the PC 102 is transmitted in a file format such as PDF, it can be viewed by a general-purpose viewer included in the PC 102.

FIG. 2 is a diagram showing a detailed configuration of the MFP 101. The MFP 101 includes a scanner unit 201 which is an image input device, a printer unit 202 which is an image output device, a control unit 204 which controls the entire MFP, an operation unit 203 which is a user interface, and the like. The control unit 204 is a controller that is connected to the scanner unit 201, the printer unit 202, and the operation unit 203, and is connected to the LAN 209 to input/output image information and device information. The CPU 205 is a processor that controls the entire system. The RAM 206 is a system work memory for the CPU 205 to operate and also an image memory for temporarily storing image data. The ROM 210 is a boot ROM, and stores programs such as a system boot program. The storage unit 211 is a non-volatile storage medium such as a hard disk drive, and stores system control software and image data. The operation unit I/F 207 is an interface unit with the operation unit (UI) 203, and outputs image data to be displayed on the operation unit 203 to the operation unit 203. Further, the operation unit I/F 207 plays a role of transmitting information instructed by the user of the image processing apparatus via the operation unit 203 to the CPU 205. The network I/F 208 connects the image processing apparatus to the LAN 209 and inputs/outputs data (for example, transmits compressed data in PDF format to another apparatus or receives compressed data in PDF format from another apparatus). Or). The above devices are arranged on the system bus 216. The image bus interface 212 is a bus bridge that connects the system bus 216 and the image bus 217 that transfers image data at high speed, and converts the data structure. The image bus 217 is composed of, for example, a PCI bus or IEEE1394. The following devices are arranged on the image bus 217. The raster image processor (RIP) 213 realizes a so-called rendering process that analyzes a PDL (Page Description Language) code and develops it into a bitmap image of a specified resolution. The device I/F unit 214 connects the scanner unit 201, which is an image input device, via a signal line 218, and the printer unit 202, which is an image output device, via a signal line 219, and synchronizes image data. Perform system/asynchronous system conversion. The data processing unit 215 generates compressed PDF data (515) by performing processing such as PDF high compression and OCR. The generated compressed data (515) is transmitted to the designated destination (for example, the client PC 102) via the network I/F 208 and the LAN 209. The data processing unit 215 can also expand the compressed data received via the network I/F 208 and the LAN 209. The decompressed image is sent to the printer unit 202 via the device I/F 214 and printed.

<Explanation of the data processing unit 215>
Next, the configuration of the image compression processing unit and the configuration of the image decompression processing unit realized by the data processing unit 215 of FIG. 2 will be described with reference to the block diagrams of FIGS. 5 and 6. The data processing unit 215 may be configured to function as each processing unit of FIG. 5 or FIG. 6 by the processor executing a computer program, or a part or all of the data processing unit 215 may be a hardware such as an ASIC or an electronic circuit. You may make it comprised by wear.

In the PDF high compression processing, as described in Patent Document 1, area determination is performed for each attribute, and binary lossless compression by MMR and multivalued lossy compression by JPEG are applied according to the attribute of each area. To change and compress. That is, the compression ratio can be increased by performing the MMR compression on the character area and the JPEG compression of the image in which the character area is filled with the surrounding color, and the high quality can be obtained for the character area. This PDF high-compression processing is a compression technique effective for color or monochrome multi-valued images. Although details will be described later, in the present embodiment, it is possible to determine whether or not the area that is destroyed when binarized is a character area. By doing so, it becomes possible to determine that only the real character area is the target to be MMR compressed.

FIG. 5 is a block diagram showing a configuration of an image compression processing unit realized by the data processing unit 215, and shows each processing unit for compressing an input image to generate a high compression PDF (with an OCR result).

The binarization unit 502 generates a binary image from the input image 501 which is a multi-valued image. In the binary image, the pixels darker than the threshold in the input image are, for example, black pixels, and the pixels below the threshold are, for example, white pixels (of course, the binarization result is not expressed in black or white, but in other colors). Or it may be represented by 1, 0 or 0, 1 with no color). Further, the binarization unit 502 aims to distinguish pixels darker than the threshold and pixels less than the threshold, but a method other than binarization may be used as long as the same purpose can be achieved (for example, 3 It may be quantized or quantized). However, the following description will be made assuming that the binarization unit 502 has performed binarization. When the input image is an image like 701, the binary image is like 702. When the input image is a color multi-valued image, binarization is performed only for the brightness of the multi-valued image (for example, Y of YUV).

The area determination unit 503 detects a character area and a photo area from the binary image generated by the binarization unit 502. Thereby, for example, 704 and 706 are detected as a character area, and 705 is detected as a photograph area. This processing is performed by a known area identification method (for example, Japanese Patent Laid-Open No. 06-068301). The outline is as follows, for example.

(1) A black pixel block (black pixel block) that continuously exists in any of eight directions is extracted by tracing the contours of black pixels that are connected to the binary image 702 by eight connections. 8-connected means that pixels of the same color (black in this case) are continuous in any of the eight directions of upper left, left, lower left, lower, lower right, right, upper right, and upper. On the other hand, 4-connected means that pixels of the same color are continuous in any of the four directions of left, bottom, right, and top.

(2) If there is a black pixel block that exceeds a certain size in the extracted black pixel blocks (for example, a black pixel block in which the area surrounded by the black pixel blocks exceeds a certain area). , Specify whether there is a white pixel block in the area. That is, a white pixel block is extracted by tracing the contours of white pixels connected by four connections in the area. Further, when the extracted white pixel block exceeds a certain size, the black pixel block is extracted by tracing the contour of the same black pixel again. These processes are repeatedly performed until the pixel block becomes a certain size or less.

(3) The obtained black pixel block is classified into a character or a photograph using at least one of size, shape, and black pixel density. For example, the aspect ratio is close to 1 (that is, it is within 1 plus or minus α. α is a fixed threshold value, for example, 0.1), and the size is determined (for example, pixels surrounded by a black pixel block). A black pixel block of 100 pixels or less) is determined as a black pixel block forming a character. Then, the remaining black pixel blocks are determined to be the pixel blocks forming the photograph.

(4) If the distance between black pixel blocks forming a character is within a predetermined distance (for example, 3 pixels), the black pixel blocks are classified into the same group. Then, the circumscribing rectangular area including all of the black pixel blocks classified into the same group is determined as the character area (704, 706). It should be noted that the black pixel blocks forming a character, which do not have other black pixel blocks forming a character within a predetermined distance, will form one group by themselves. Therefore, the circumscribed rectangular area of the single black pixel block is determined as the character area. It should be noted that the same processing as the processing described in (4) is performed on the black pixel block forming the photograph.

(5) The position of each area and the attribute judgment information (character or photograph) of the area are output as the judgment result.

By the above processes (1) to (5), the determination result that 704 and 706 are the character region and 705 is the photograph region is output. This is the end of the description of the area determination unit 503.

The character cutout unit 504 performs a character cutting rectangle cutout process on each of the character areas generated by the area determination unit 503. The cut out results are like 710, 711, 712, 713. This cutout process is composed of the following processes.

(1) Select one of the character areas (for example, select 708).

(2) A projection is taken from the lateral direction with respect to one binary image specified by the character area. Specifically, the number of black pixels in a line extending in the horizontal direction is counted, and the counted result is a projection. The projection taken is represented at 715. In this projection 715, lines that continue in the vertical direction and have more black pixels than the threshold value are grouped together. This results in three groups. The three groups are a group composed of line groups in which ABCD exists, a group composed of line groups in which EFG exists, and a group composed of line groups in which H exists.

(3) For each group, take a projection from the vertical direction. Reference numeral 716 represents a projection taken on a line group in which ABCD exists.

(4) In the projection of each group, horizontally continuous lines having more black pixels than the threshold value are grouped into one group. For example, projection 716 will occur in four groups. The four groups are composed of a group consisting of line groups with A, a group consisting of line groups with B, a group consisting of line groups with C, and a line group containing D. It consists of a group.

(5) The circumscribing rectangle of the group of each line group obtained in (4) is cut out as a character cutting rectangle. As a result, for example, the circumscribed rectangle of each character is cut out as a character cutout rectangle. The cut-out results are as shown in 711, 712, 713, and 710.

(6) The above processes (1) to (5) are repeated until there are no unselected character areas.

Here, an example of an image to be processed and an image of a processing result of binarization/region determination/character cutout will be described with reference to FIG. 7. An image 701 is an example of the input image 501, 7011 is a character image described on a white background, 7012 is a character image described on a light density background, and 7013 is a character described on a dark density background. The example of an image is shown. That is, the characters 7011 and 7012 are character images with a large density difference between the character and the background, and the character image 7013 is a character image with a small density difference between the character and the background.

Reference numeral 702 denotes an example of a binary image obtained by binarizing the image 701 in the binarizing unit 502. The character 7013 is binarized with a threshold value smaller than the density of the background, and is blackened. It shows that it has been exhausted.

In the present embodiment, a character image that is crushed when binarized (for example, a darker character on a background darker than a threshold or the like has a small density difference between the background and the character, and the background and the character are separated even when binarized. A difficult image) is called a “character image that is difficult to separate from the background”. In addition, a character image that does not collapse when binarized (for example, white or black characters on a background that is lighter than a threshold value, is easy to separate the background and the character when binarized due to a large density difference between the background and the character. Image) is referred to as a “character image that is easily separated from the background”. That is, the "character image that is easily separated from the background" is an image of a character area in which the character image portion becomes black pixels when binarized and the background portion other than the character becomes white pixels.

Reference numeral 703 denotes the result of the region determination unit 503 performing the region determination on the binary image 702. As a result of the area determination, it is assumed that 704 and 706 are determined to be character areas and 705 is determined to be a photograph area. The character areas 707 and 708 are obtained by extracting, from the binary image 703, partial images determined as the character areas by the area determination unit 503. Reference numeral 709 shows a schematic view of the character cutting rectangle cut out by the character cutting unit 504. A character cutting rectangle 710 is cut out from the character area 704. Further, 711, 712, and 713 are character cutting rectangles cut out from the character area 706.

The area determination unit 2 (505) determines whether or not the character image in the character cutting rectangle cut out by the character cutting unit 504 is a character that is crushed when binarized (a character image that is difficult to separate from the background). To do. Details of the determination method of the area determination unit 2 will be described later. The character region information generated by the region determination unit 503 and the character segmentation unit 504 based on the information of the character region determined by the region determination unit 2 (505) to be a “character image that is difficult to separate from the background”. The character cutting rectangle information that has been corrected is corrected. That is, from the character area information generated by the area determination unit 503 and the character cut rectangle information generated by the character cutout unit 504, the area determination unit 2 (505) determines that it is a “character image that is difficult to separate from the background”. Information on the determined character area is removed. By doing so, the character area determined to be a “character image that is difficult to separate from the background” is determined not to be a character, so that the MMR compression described later does not apply and the character image becomes invisible. That will solve the problem.

The MMR compressing unit 506 extracts a binary image of a character area from the binary image generated by the binarizing unit 502 based on the character area information corrected by the area determining unit 2 (505) (that is, Only the binary image included in the character-cutting rectangular area determined to be “a character image that is easily separated from the background” is extracted). Then, MMR compression is performed on the extracted binary image of the character region to generate compression code 1 (511).

The reduction unit 507 performs a reduction process (resolution reduction process) on the input image 501 to generate a reduced multivalued image (not shown).

The representative color extraction unit 508 identifies the position of the pixel (black pixel) forming each character in the binary image based on the character area information and the character cutting rectangle information after being corrected by the area determination unit 2 (505). To do. Then, based on the position of the pixel of the identified character, the color of the corresponding position in the reduced multi-valued image is referred to, the representative color of the character is calculated for each character cutting rectangular area, and the character color information 513 of each character is calculated. To get For example, the representative color is an average or weighted average of colors in a multi-valued image of a pixel group that is black in the binary image in the character-cut rectangular area. Alternatively, it is the most frequent color in such a pixel group. As described above, there are various possible ways of obtaining the representative color, but the color in the multi-valued image of at least one pixel of the pixel group that becomes black in the binary image in the character-cutting rectangular area is used for the calculation of the representative job. Will be used.

The character area filling unit 509 identifies the position of a pixel (black pixel) forming each character in the binary image based on the character area information and the character cutting rectangle information after being corrected by the area determination unit 2 (505). To do. Then, based on the position of the identified pixel, the pixel at the corresponding position in the reduced multi-valued image is filled with the surrounding color. As the peripheral color, an average value of pixel values of pixels around the character may be used, and the pixel value of the pixel of the character may be replaced with the obtained peripheral color. The details of the filling processing by the character area filling portion are described in Patent Document 1.

The JPEG compression unit 510 JPEG-compresses the image that has undergone the padding processing by the character area padding unit 509 to generate a compression code 2 (514).

The OCR unit (516) performs known character recognition processing on the area determined by the area determination unit (503) as the character cutting rectangle information generated in step 904. The character code 517 is a character code obtained by the character recognition process.

Here, when MMR compression is performed by the MMR compression unit (506), the region determined by the region determination unit 2 (505), that is, the region determined as “a character image that is easily separated from the background” is targeted. In contrast to the MMR compression as described above, when the OCR unit (516) performs the OCR, the region determined by the region determination unit (503) as the character region is subjected to the OCR.

Of these, the “character image that is easily separated from the background” is a part of the area that is determined to be a character area by the area determination unit (503). Narrower. That is, the OCR target area is wide and the MMR compression area is narrow.

Why is the OCR area wider? This is because even if there is something that is not really a character in the OCR target area, you will only get an extra character code and it will not cause a big problem (if you think that it is an extra character, Erase the character code). On the other hand, when the MMR compression is performed on a region that is not actually a character, the image quality of the region deteriorates. Therefore, in OCR, processing is performed for a wide area, and for MMR compression, processing is performed for a narrow area.

In this way, the compressed code 1 (511) obtained from each component, the corrected character area information (512), the character color information (513), the compressed code 2 (514), and the character code ( A file of compressed data (515) including 517) is generated in PDF format. The generated PDF format file will be transmitted to the destination designated by the user as described above.

FIG. 6 is a block diagram showing the configuration of an image expansion processing unit that expands PDF format compressed data sent from another device. The process of FIG. 6 is executed when decompressing compressed data and printing. Here, a case where the compressed data sent from another device is the same file as 515 will be described as an example.

The MMR expansion unit 601 performs MMR expansion processing on the compression code 1 (511) included in the file of the compressed data (515) to reproduce a binary image. The JPEG decompression unit 603 performs JPEG decompression processing on the compressed code 2 (514) to reproduce a reduced multi-valued image. The enlarging unit 604 performs an enlarging process on the reduced multi-valued image decompressed by the JPEG decompressing unit (603) to generate a multi-valued image having the same size as the uncompressed input image 501.

The synthesizing unit 602 assigns a color (hereinafter, referred to as a character color, 513) of the character color information to the black pixel of the binary image expanded by the MMR expanding unit with reference to the character area information (512). Further, the binary image to which the character color is assigned is combined with the multi-valued image generated by the enlargement unit 604 to generate a decompressed image 605. When synthesizing, a transparent color is assigned to the white pixel in the binary image, and the background multi-valued image is transparent. In this way, the image decompression processing unit decompresses the compressed data generated by the image compression processing unit and generates a decompressed image 605. The decompressed image 605 is sent to the printer unit 202 via the device I/F 214 and printed. The image expansion processing unit ignores the character code 517. This is because the character code is unnecessary for printing the expanded image. A device such as the client PC 102 that displays the decompressed image 605 on the display needs a character code, not the MFP 101. Therefore, the MFP 101 ignores the character code 517. To be exact, it is not the PC 102 that needs the character code but the user who uses the PC 102. The character code is used when you want to cut and paste a character string and edit it.

Next, details of the processing executed by the above-described area determination unit 2 (505) will be described. The area determination unit 2 (505) uses the binary image generated by the binarization unit 502, the reduced multi-valued image generated by the reduction unit 507, and the character cutting rectangle information generated by the character cutting unit 504. Based on this, it is determined whether the character image in the character cutting rectangle is destroyed by binarization. When the input image 501 is held in the storage unit 211, the input image 501 may be used instead of the reduced multi-valued image.

The detailed configuration of the area determination unit 2 (505) will be described with reference to FIG. In describing, the example of the character image of FIG. 4 will be referred to as appropriate. Here, 401 in FIG. 4 shows an example of a character (an image in which the density difference between the background and the character is large) written on a white background. When the binarizing unit 502 binarizes the image 401, A binary image as shown by 402 is obtained. Reference numeral 406 denotes an example of a character (an image in which the density difference between the background and the character is small) written on a background having a high density, and the image 406 is binarized with a threshold value smaller than the density of the background 407. Then, an image is crushed in black as shown by 408. It should be noted that the characters written on the background of light density become the same as the image 402 by being binarized with a threshold value between the background density and the character density, and therefore description thereof is omitted. 403-405 and 409-411 are mentioned later.

The area determination unit 2 (505) includes a thinning unit 301, an edge detection unit 302, a logical operation unit 303, an edge count unit 304, and an edge number comparison unit 305.

The area determination unit 2 (505) extracts the edge pixel inside the area darker than the threshold value (that is, the area darkened at 402 and 408) (1). Then, when the number of extracted edge pixels is less than the threshold value, it is determined to be a “character image that is easily separated from the background” (2). Further, when it is equal to or more than the threshold value, it is determined that the character image is a character image that is difficult to separate from the background (2).

For example, there are no edge pixels inside the blackened area 402. On the other hand, inside the blackened area 408, there are edge pixels (H edge pixels represented by 410). The edge pixel referred to here does not mean the edge pixel extracted from the binary image, but the edge pixel extracted from the multi-valued image (input image).

The following configuration is one configuration for realizing the above processes (1) and (2), and is not limited to this configuration. Other possible configurations will be described later.

The thinning unit 301 performs a thinning process on a character-cutting rectangle unit basis with reference to the character-cutting rectangle information for a binary image. The thinning process thins the black pixel block by removing two pixels outside the black pixel block in the binary image (that is, replacing the black pixel on the outline of the black pixel block with the white pixel). This is processing for. For example, each pixel in the binary image included in one target character cutting rectangle is sequentially set as a target pixel, and scanning is performed using a 5×5 window. Then, if there is even one white pixel in the 5×5 window, the thinning process is performed by replacing the pixel of interest (center of 5×5) with the white pixel. Here, if the thinning process is performed on the binary image 402, a thinned image like 403 is obtained. Further, if the binary image 408 is thinned, a thinned image like 409 is obtained.

The edge detection unit 302 performs edge detection on the input reduced multi-valued image in units of character cutting rectangles with reference to the character cutting rectangle information. An image in which a pixel determined to be an edge is represented as a black pixel and a pixel determined not to be an edge is represented as a white pixel is referred to as an edge detection image. Since a known method may be used for the edge detection, the details thereof will be omitted, but the following processing can be considered. For example, the differential filter processing is performed on the luminance component of the reduced multi-valued image to obtain the edge strength of each pixel, the pixels whose edge strength is equal to or higher than a predetermined threshold are black pixels, and the pixels whose edge strength is lower than the predetermined threshold are selected. An edge detection image is generated by using white pixels. However, more accurate edge detection can be realized by using the edge detection method described in the fourth embodiment. When edge detection is performed on a reduced multivalued image (not shown) obtained by reducing the input image 401, an edge detection image such as 404 is obtained. If edge detection is performed on a reduced multi-valued image (not shown) obtained by reducing the input image 406, an edge detection image like 410 is obtained. Here, when the reduced multi-valued image (not shown) obtained by reducing the input images 401 and 406 has a half resolution of the input image, 404 and 410 also have a half resolution of the input image. However, for simplification of description, they are illustrated in the same size. When the storage unit 211 holds the input images 401 and 406, the edge detection may be performed using the input images 401 and 406 instead of the reduced multivalued image.

The logical operation unit 303 takes a logical product (AND) of the thinned image generated by the thinning unit 301 and the edge detection image generated by the edge detection unit 302 to generate a logical product (AND) image. Perform processing. Specifically, the logical product is calculated only when the thinned image generated by the thinning unit 301 has a black pixel and the black pixel exists at the same position in the edge detection image generated by the edge detection unit 302. It becomes a black pixel. If the edge detection image generated by the edge detection unit 302 has a resolution half that of the thinned image, the edge detection image is subjected to zero-order interpolation to match the resolution of the thinned image, and then the logical product is calculated. To take. Alternatively, the thinned images are thinned out to obtain the logical product after matching the resolution of the edge detection image. When the logical product of the thinned image 403 and the edge detection image 404 is calculated, the black pixels of the thinned image 403 and the black pixels of the edge detection image 404 are not at the same position, and therefore the black pixel in the logical product image 405 is basically (However, it may remain a little due to the influence of noise etc.). On the other hand, when the logical product of the thinned image 409 and the edge detection image 410 is obtained, black pixels remain in the outline portion of the character like the logical product image 411. As described above, the number of black pixels in the logical product image for the “character image that is easily separated from the background” is small, and the number of black pixels in the logical product image for the “character image that is difficult to separate from the background” is large. ..

412 is a diagram showing a case where the thinned image 403 and the edge detection image 404 are superimposed. 413 corresponds to a black pixel of the thinned image 403, 414 corresponds to a black pixel of the edge detection image 404, and the black pixel of the thinned image 413 and the black pixel of the edge detection image 414 are not at the same position. , If a logical product is taken, a black pixel will not be generated.

The edge counting unit 304 performs a process of counting, as the number of edges, the number of black pixels in the result (logical product image) obtained by taking the logical product (AND) by the logical operation unit 303.

The edge number comparing unit 305 compares the number of edges counted by the edge counting unit 304 with a predetermined threshold value to determine whether the image is a “character image that is easily separated from the background” or a “character image that is difficult to separate from the background”. To judge. That is, if the number of edges is less than a predetermined threshold value, it is determined that the character image is a character image that is easily separated from the background (a character image that is not crushed when binarized), and if the number of edges is equal to or greater than the predetermined threshold value, the “background Is a character image that is difficult to separate (a character image that is crushed when binarized)”.

If the width of the pixels of the black pixel block is smaller than the narrowing width of the thinning process, the black pixel block may disappear entirely in the binary image due to the thinning process. For example, if a black pixel block of a binary image is a thin line character having a width of 3 pixels and the narrowing width of the thinning process is 4 pixels, the black pixel block disappears when the binary image is thinned. End up. When the black pixel block disappears in this way, it is preferable to omit the processes of the edge detection unit 302, the logical operation unit 303, the edge count unit 304, and the edge number comparison unit 305 from the viewpoint of improving the processing speed. This is because even if the edge detection unit 302 detects an edge pixel, it is clear that the count result becomes 0 when the logical product is calculated with the thinned image and the number of edges obtained as a result is counted. .. When the count result is 0, it means that the number of edges is less than the predetermined threshold value, and thus it can be determined that the character image is a character image that is easily separated from the background (character image that is not crushed when binarized). Therefore, when all the black pixels in the target character cutting rectangle are eliminated by the thinning processing, the processing of the edge detecting unit 302-the edge number comparing unit 305 is not performed, and the character cutting rectangle is set as the “background”. Is a character image that can be easily separated (a character image that does not collapse when binarized)”. When the processing of 302 to 305 is omitted in this way, the processing advances to the processing of the thinning unit 301-edge number comparison unit 305 for the next character cutting rectangular area. The reason why the above processing is omitted can be explained as follows. That is, if the black pixels disappear after thinning, it can be said that the black pixel group of the original binary image is quite thin, but the thin black pixel group is generally a character or a line. Therefore, the processing speed is determined by omitting the above processing and determining that the target character cutting rectangular area is a “character image that is easily separated from the background (character image that does not collapse when binarized)”. Can be explained as being preferable.

Alternatively, when all the black pixels in the binary image are lost, the number of pixels to be cut can be reduced. For example, if there is at least one white pixel in a 5×5 window and the target pixel (5×5 center) is replaced with a white pixel, the black pixel block becomes all white pixels. It is also possible to make the value smaller and process in a 3×3 window. Note that thinning and thinning processing have the same meaning.

In the above description, it is described that the number of edges counted by the edge counting unit 304 is compared with a predetermined threshold value. However, a value obtained by dividing the number of edges by the number of black pixels of the thinned image is a predetermined threshold value. It is also preferable to compare. By doing so, an appropriate judgment can be made regardless of the size of the character cutting rectangular area. It is also possible to divide the number of edges by the total number of pixels forming the character-cut rectangular area or the number of black pixels after binarizing the rectangular area. However, the highest accuracy is obtained by dividing the number of edges by the number of black pixels of the thinned image, as described above. This is because it is possible to know at what rate the edges exist inside the binary image (inside the dark area). It can be said that the higher this ratio is, the higher the ratio of the edges existing inside the binary image is, and thus the higher the possibility that the binary image is not a character.

Next, each processing executed by the data processing unit 215 will be described using the flowchart of FIG. In describing, reference will be made to FIGS. The area determination unit 2 (505) executes the processing of 905 to 911 in FIG. 8.

In step 901, the binarization unit 502 executes binarization processing on the input image 501.

In step 902, the area determination unit 503 performs area determination processing on the binary image, identifies each area included in the binary image, and determines whether the identified area is a character area or a non-character area. Is determined.

In step 903, one of the areas determined by the area determination unit is sequentially set as the attention area, and when the attention area is the area determined by the area determination unit as the character area, the process proceeds to step 904 and the non-character area is selected. If the area is determined to be, the process proceeds to step 913.

In step 904, the character cutout unit 504 generates character cut rectangle information by performing character cutout on the image in the attention area.

In step 916, the OCR unit 516 performs known character recognition processing on the area determined to be the character area by the area determination unit (503) with reference to the character cutting rectangle information generated in step 904. ..

In step 905, the thinning unit 301 refers to the binarized image binarized in step 902 with respect to the binarized image in step 904 and refers to each binarized image in the binarized rectangle. The thinning process is executed.

In step 906, the edge detection unit 302 uses the reduced multivalued image (or the input image 501) obtained by reducing the input image and the character-cutting rectangle information generated in step 904 to reduce the size of the character-cutting rectangle. The edge detection processing is executed for each value image (or the input image within the character cutting rectangle).

In step 907, the logical operation unit 303 takes the logical product (AND) of the thinned image generated by the thinning unit 301 in step 905 and the edge image generated in step 906.

In step 908, the edge counting unit 304 counts the number of black pixels in the logical product image resulting from the logical product (AND) in step 907 and obtains the number of edges. Here, the calculated number of edges may be further normalized by dividing by the area of the character-cut rectangular area (the total number of pixels in the character-cut rectangular area) to obtain the number of edges per unit area. .. This has the advantage that the threshold value can be compared in step 909 without depending on the size of the character-cutting rectangular area.

Next, in step 909, the edge number comparison unit 305 compares the edge number counted in step 908 with the threshold th. Here, when the number of edges is larger than the threshold value th, it is determined in step 910 that the target character cutting rectangular area is a “character image that is difficult to separate from the background”. If the number of edges is less than or equal to the threshold value th, in step 911 the target character-cutting rectangular area is determined to be a “character image that can be easily separated from the background”.

In step 912, the character cutout unit 504 determines whether or not the processing has been completed for all the character cut rectangles in the focused character area, and if it is determined that the processing has been completed, the process proceeds to step 913. On the other hand, if it is determined that there is an unprocessed character cutting rectangle, the next character cutting rectangle is set as the processing target in step 914, and the process returns to step 905.

If it is determined in step 913 that the determination has been completed for all areas, this processing ends, and if it is determined that there is an unprocessed area, the next unprocessed area is set as the attention area in step 915. Then, the process returns to step 903.

As described above, in the area determination unit 2 (505), based on the number of black pixels (the number of remaining edges) obtained as the logical product of the thinned image and the edge detection image for each character-cut rectangular area. It becomes possible to determine with high accuracy whether each character-cutting rectangular area is a “character image that is easily separated from the background” or a “character image that is difficult to separate from the background”.

The “character image that is difficult to separate from the background” (for example, 713 in FIG. 7) is removed from the character area information, and thus is not a processing target of the MMR compression unit 506. That is, the “character image that is difficult to separate from the background” is not binarized but is compressed by the JPEG compression unit 510 together with the background image.

As described above, it is possible to determine whether or not the character image is crushed when binarized. Therefore, when the PDF high compression is applied, the character image can be prevented from being crushed.

In the present embodiment, as shown in 406 and 408 of FIG. 4, the case of one character "H" is defined as "a character image that is difficult to separate from the background (a character image that is crushed when binarized)". However, the present invention is not limited to this, and may be two or more characters as shown in the input image 1001 of FIG. 9, for example. It should be noted that when this input image 1001 is binarized, it becomes a binary image 1002. Further, the character image crushed by binarization does not need to be rectangular, and may be an image in which a part of the character image is crushed as shown in 1003 of FIG. 9, for example. In addition, when this input image 1003 is binarized, it becomes a binary image 1004.

Next, another configuration of the area determination unit 2 (505) will be described.

In another configuration, (A) first, the image input to the area determination unit 2 is divided into an area darker than a threshold and an area equal to or less than the threshold (binarization, ternarization, or another method may be used). As a result, areas such as 402 and 408 are available.

Then, (B) in the input image, edge pixels are extracted from a region determined to be darker than a threshold value (H region of 401 or entire region of 406) (the extraction method is as described above). When this edge is extracted, the end portion of the area determined to be darker than the threshold value (for example, a pixel within one pixel or a pixel within two pixels from the edge) is not a target. That is, in the configuration of B, only the edge pixels that are separated by a certain distance or more from the end portion of the area determined to be darker than the threshold value (entered inside) are extracted. Alternatively, a configuration may be adopted in which edge pixels are extracted for such edge portions (pixels that are not separated by a certain distance or more) and the edge portions are removed. By doing so, the same result as that of 405 or 411 is obtained. In this example, the fixed distance is 3 pixels, but it may be another value.

After (C), the number of edge pixels at the obtained result is counted, and it is determined whether the number of edge pixels is larger than the threshold value th or less than the threshold value th.

By doing so, the same result as the above-described method (a determination result as to whether it is a “character image that is difficult to separate from the background” or a “character image that is easy to separate from the background”) is obtained. Instead of the process (B), edge pixels may be extracted from the entire image input to the area determination unit 2. In that case, among the edge pixels extracted from the entire input image, the edge part of the area determined to be darker than the threshold and the area equal to or less than the threshold are excluded. By doing so, the same result as the above-mentioned configuration of (B) can be obtained.

In this embodiment, the processing of the character cutout 904 cut out by the character cutout unit 504 on a character-by-character basis has been described. This processing can be performed by further dividing the character unit instead of the character unit. For example, it is possible to divide the area into four equal parts for the character cutout unit 504 and perform the processing in each area. For example, 1300 to 1304 in FIG. 16 are examples in which the character-cut 406 is equally divided into four. Processing is performed in each of 1300 to 1304. Further, the determination can be made only in the central portion of the character-cut out area (for example, only 60% of the central portion of the character-cut area is used) instead of dividing the character evenly. For example, 1305 in FIG. 16 is obtained by extracting 60% of the central portion of the character-cut 406, and processing is performed on this 1305. In addition, the judgment in the area where the character is cut out and the judgment in the area division and/or the central portion are combined to be the “character image that is easily separated from the background” or the “character image that is difficult to separate from the background”. It is also possible to determine whether or not.

(Example 2)
In the first embodiment, the area determined by the area determination unit 2 (505) as a “character image that is difficult to separate from the background” is not subjected to MMR compression processing. In the second embodiment, the region determination unit 2 (505) again performs the high-precision binarization process on the region determined to be “a character image that is difficult to separate from the background” using an algorithm different from that of the binarization unit 502. Then, the pixels of the character image portion may be separated from the background. In this case, the image quality of the character area can be improved by performing the MMR compression processing using the character area resulting from the highly accurate re-binarization processing. For example, since the area 713 in FIG. 7 is determined to be a “character image that is difficult to separate from the background”, only the area 7013 in the input image 701 corresponding to the area 713 is binarized with a threshold different from other areas. To do. As a result, a binary image as shown by 714 in FIG. 7 can be generated, and this character area can be MMR compressed. An example of the highly accurate re-binarization process is a method of performing the binarization process using the average value of the density or the brightness of the target region as the threshold value, instead of performing the binarization process with the fixed threshold value. ..

(Example 3)
In the first embodiment, an input image having relatively good character quality as shown by 401 in FIG. 4 is taken as an example. However, when the edge detection processing is performed on an image (for example, a scanned document or a compressed image) that is poor in character quality and has a lot of noise as shown by 1101 in FIG. 10, inside the character as shown by 1102 in FIG. Many edges may appear. The appearance of edges inside a character tends to be more prominent as the character becomes larger.

Here, in the logical product (AND) image 1104 obtained from the edge detection image 1102 and the thinned image 1103, the edge inside the character tends to remain. If many edges inside the character remain, the character image is determined to be a “character image that is difficult to separate from the background” although it is originally a “character image that is easy to separate from the background”.

In the third embodiment, when the size of the character cutting area is large, the amount of reduction of the thinning processing in the thinning unit 301 is increased, so that the edge left inside the character can be reduced. This process will be described with reference to 1105-1112 in FIG.

Reference numeral 1105 denotes a lowercase character of the input image that is poor in character quality and has a lot of noise. Reference numeral 1106 denotes an edge detection image obtained as a result of executing the edge detection processing on the lower case image 1105. Reference numeral 1107 denotes a thinned image obtained as a result of performing the thinning processing on the lowercase image 1105. In the thinning process, a 5×5 window is used to replace the target pixel (center of 5×5) with a white pixel if there is even one white pixel in 5×5. There is.

Reference numeral 1108 denotes a logical product (AND) image of the logical product of the edge detection image 1106 and the thinned image 1107. Here, even if the input image is in lower case with poor character quality and much noise, the number of edges is smaller than in the upper case logical product (AND) image 1104.

Further, reference numeral 1109 denotes an uppercase character (character image similar to 1101) of the input image having poor character quality and having many noises. Reference numeral 1110 denotes an edge detection image obtained as a result of performing the edge detection processing on the capital image 1109. Reference numeral 1111 denotes a thinned image obtained as a result of performing the thinning processing on the capital image 1109. In the thinning process for an uppercase image, a 9×9 window is used, and if there is even one white pixel in 9×9, the pixel of interest (center of 9×9) is replaced with a white pixel. Perform processing. That is, the reduction amount of the thinning process is increased by changing the size of the window based on the size of the character image (size of the character cutting area). The size of the window described above is an example, and is not limited to 5×5 or 9×9.

Reference numeral 1112 denotes a logical product (AND) image of the logical product of the edge detection image 1110 and the thinned image 1111. The logical product (AND) image 1112 has a smaller number of edges than the logical product (AND) image 1104 described above. Therefore, even if it is a capital letter with a lot of noise, if the size of the character image is large, it is possible to determine that the character image is “easily separated from the background” by increasing the reduction amount of the thinning process.

As described above, according to the third embodiment, by controlling the reduction amount of the thinning processing by the thinning unit based on the size of the character cutting region, the case where the input image is a scanned document can be realized. Also, the influence of noise or the like can be reduced, and highly accurate determination can be performed.

(Example 4)
Next, the details of the processing performed by the edge detection unit (302) in the area determination unit 2 (505) in FIG. 5 will be described using FIG. 11. The edge detection unit (302) includes a variance value detection unit 1001, an edge determination threshold value calculation unit 1002, and an edge extraction unit 1003. In order to describe the processing of the edge detection unit (302) in more detail, the description will be given with reference to FIG. 12. Reference numerals 1101, 1102, and 1103 in FIG. 12 denote input images cut out in character cutting rectangle units with reference to the character cutting rectangle information for the input image similarly to 401 and 406 shown in FIG. Reference numerals 1101, 1102, and 1103 are image examples in which the signal values obtained by the scanner unit 201 are different. To be more specific, the signal values in the L*a*b* color system are shown, where L* is the lightness and a* and b* are the chromaticities. In this example, the L*a*b* color system is shown, but the present invention is not limited to this, and similar processing can be performed with signal values in another color space such as the RGB color system. The signal value of the area indicated by 1104 of 1101 is {L*, a*, b*}={1128, -50, +30}. The signal value of the area indicated by 1105 is {L*, a*, b*}={128, +50, -60}. An example in which there is a large signal value difference between the areas 1104 and 1105 is shown. On the other hand, the signal value of the area indicated by 1106 of 1102 is {L*, a*, b*}={128, -50, +30}. The signal value of the area indicated by 1107 is {L*, a*, b*}={128, -60, +30}. An example in which there is only a small signal value difference between the areas 1106 and 1107 is shown. Further, the signal value of the area indicated by 1108 of 1103 is {L*, a*, b*}={128, -50, +30}. The signal value of the area indicated by 1109 is {L*, a*, b*}={128, -52, +30}. An example in which there is almost no signal value difference between the areas 1108 and 1109 is shown. For example, when the edge detection unit (302) is not configured as described above and edge detection is performed simply based on signal value comparison with adjacent pixels or edge detection performed by filter processing, there are the following problems. That is, depending on the threshold value, the contour edge can be acquired at the boundary between 1104 and 1105 in 1101, but the contour edge cannot be acquired at the boundary between 1106 and 1107 at 1102. Further, when the threshold is set such that the contour edge can be acquired at the boundary between 1102 and 1107 in 1102, the contour edge at the boundary between 1108 and 1109 in 1103 is acquired. As a result, small noises such as scanner reading variations and Jpeg noises are also detected as edges.

A configuration for solving the above problem is shown in FIG. 11, and the variance value detection unit 1001 is a computation unit that computes a variance value in the signal value of the input image cut out in character cutting rectangle units. The calculation method is, for example, the following formula.

Here, the number of pixels of the clipped input image is n, and the signal value of each pixel (in this embodiment, the respective values of L*, a*, and b*) is Xi (i=1, 2,..., N). ), the average of the signal values of the number of pixels in the area is indicated by Xave. In the present embodiment, the dispersion value at each value of L*, a*, and b* is shown, but it is not limited, and for example, the covariance value at the a* and b* signal values may be used. Good. In the examples 1101, 1102, and 1103 shown in FIG. 12, 1101 has a large difference in signal value and thus a large variance value, and 1102 and 1103 have a small difference in signal value and thus have a relatively small variance value.

As a definition of terms used in the following description, “a threshold value at which an edge is easily acquired” is compared with a signal value difference between adjacent pixels, and if there is a difference, in the process of determining an edge, the signal value difference is small. Is also determined as an edge. On the other hand, the “threshold value at which an edge is hard to be acquired” refers to an edge that is not determined as an edge unless the signal value difference is large and is not determined as an edge when the signal value difference is small.

The edge determination threshold value calculation unit 1002 calculates a threshold value for edge extraction based on the variance value calculated by the variance value detection unit 1001. For example, as shown by 1101, a threshold value with which an edge is hard to be acquired is assigned to an image having a large variance value. On the other hand, to 1102 and 1103, a threshold value with which an edge is easily acquired is assigned.

The edge extraction unit 1003 is a processing unit that performs edge extraction processing based on the threshold determined by the edge determination threshold calculation unit 1002. The processing method may be general-purpose processing, for example, by comparing the signal value differences between adjacent pixels and determining whether or not the difference exceeds a specific threshold, or by using a filter that calculates the first derivative There is a method of obtaining the amount and making a determination based on whether or not the amount exceeds a specific threshold value.

In the case of cutting according to the condition calculated by the edge determination threshold value calculation unit 1002, 1101 assigns a threshold value at which an edge is hard to be acquired and performs edge extraction. Here, for example, an example in which the threshold determined based on the variance value is 5 is shown. When the determination is made with the threshold value, the signal value difference between the regions 1104 and 1105 is large, and therefore the edge between the regions 1104 and 1105 can be accurately extracted. The result is shown in 1110. On the other hand, in the case of 1102, although the signal value difference between 1106 and 1107 is small, an edge between the regions of 1106 and 1107 can be extracted by assigning a threshold value with which an edge is easily acquired. The result is shown in 1111. In the case of 1103, a threshold value with which an edge is easily acquired is assigned, but the signal value difference between 1108 and 1109 is much smaller than the signal value difference between 1106 and 1107. Therefore, even if the edge is a threshold value that is easily acquired, the edge between the areas 1108 and 1109 is not extracted. The result is shown in 1112.

Next, the edge detection unit (302) in FIG. 11 will be described using the flowchart in FIG. 11 will be appropriately referred to in the description.

First, in step 1201, the variance value calculation unit (1001) calculates the variance value of a signal for the input image (501). At this time, if the number of channels of the image is three, all three may be obtained, or one channel may be obtained.

Next, in step 1202, the edge threshold calculation unit (1002) determines whether or not the variance value of the image signal calculated in step 1201 exceeds a predetermined value. If it is equal to or larger than the predetermined threshold value, “a threshold value at which an edge is easily acquired” is acquired at 1203. On the other hand, if it is less than the predetermined threshold value, “a value at which an edge is hard to be acquired” is acquired at 1204.

Finally, in step 1205, the edge extraction unit (1003) performs edge extraction processing based on the threshold determined in 1203 or 1204.

As described above, in the present embodiment, when the edge extraction is performed, the threshold is adaptively switched for each input image cut out in the character cutting rectangle unit based on the variance value of the image. By doing so, it becomes possible to more accurately separate the “character image that is difficult to separate from the background” and the “character image that is easy to separate from the background” with high accuracy.

(Example 5)
The fourth embodiment has described the method of switching the threshold value based on the variance value of the signal values in the threshold value calculation when the edge is extracted. When the input image is a color image having three channels or the like, it is possible to calculate as many variance values as the number of channels, which can be used for determining the threshold value with high accuracy. However, when the input image is grayscale, since the number of channels is one, the variance value that can be used for threshold calculation is one, and it is difficult to calculate the threshold with high accuracy.

Therefore, in this embodiment, as shown in FIG. 14, the configuration of the edge detection unit (302) includes a variance value detection unit 1001, an edge determination threshold value calculation unit 1002, an edge extraction unit 1003, and a black pixel density calculation unit 1004. To be done. In addition to the input image, it is used as a binarized image.

The black pixel density calculation unit 1004 is a calculation unit that calculates the ratio of the number of black pixels to the area of the character cutting rectangle based on the input binarized image. In the input binary image, the number of black pixels is counted, and the counted number is divided by the area of the character-cutting rectangle.

Next, the edge threshold calculator 1002 calculates an optimum threshold based on the black pixel density calculated by the black pixel density calculator 1004. Here, the edge threshold is calculated according to the black pixel density in the same manner as the edge threshold is switched according to the variance value in the first embodiment. Specifically, when the black pixel density is high, the "threshold is easily acquired" is set, and when the black pixel density is low, the "edge is hard to be acquired" is set. By setting in this way, in the case of "characters having a dark background", the black pixel density is high, and the "threshold value at which an edge is easily acquired" can be used for edge extraction, and the edge can be accurately calculated. It will be possible.

Note that it is possible to use either one of the threshold value calculated based on the variance value and the threshold value calculated based on the black pixel density, but it is also possible to use both of them for the threshold value calculation. In that case, it is preferable to use the “threshold value at which edges are easily acquired” from the viewpoint of acquiring more edges, but it is also possible to select the “threshold value at which edges are hard to be acquired”. Further, by switching the weights of the respective threshold values, it is possible to give priority to the threshold value calculated based on the variance value, for example.

Further, as shown in FIG. 15, the configuration of the edge detection unit (302) includes a variance value detection unit 1001, an edge determination threshold value calculation unit 1002, an edge extraction unit 1003, a black pixel density calculation unit 1004, and a closed loop number calculation unit 1005. It may be configured.

The closed loop number calculation unit 1005 is a calculation unit that performs a labeling process for calculating the number of closed loops formed by continuous pixels in the white part of the input binarized image.

Next, the edge threshold value calculation unit 1002 calculates an optimum threshold value based on the closed loop number calculated by the closed loop number calculation unit 1005. Here, as in the first embodiment, the threshold used for edge extraction is calculated depending on the number of closed loops. Specifically, when the number of closed loops is large, the “threshold value at which an edge is hard to be acquired” is used, and when the number of closed loops is small, a “threshold value at which an edge is easily acquired” is used.

With the above processing, it is possible to calculate the optimum edge threshold even for an image such as gray scale in which the number of channels is small and the edge threshold cannot be calculated based on the variance of signal values.

(Other embodiments)
The present invention supplies a program that implements one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium, and one or more processors in a computer of the system or apparatus read and execute the program. It can also be realized by the processing. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

Claims (8)

First determining means for determining whether the attribute of the area in the image is at least a character attribute or an image attribute based on the binarization result of the image ;
Second determining means for determining whether the image of the area determined by the first determining means to have the character attribute is a character image easily separated from the background or a character image difficult to separate from the background ,
When the image of the area determined by the first determination means to have the character attribute is a character image that is easily separated from the background by the second determination means, binary compression processing is performed on the image of the area. When the image of the area determined by the first determining means to have the character attribute is a character image that is difficult to separate from the background by the second determining means, the image of the area is displayed . Processing means for executing multi-value compression processing,
A file of the image is generated based on at least the image of the area that has been determined by the first determination unit to have the character attribute, and one of the binary compression processing and the multi-value compression processing has been executed. An image processing apparatus comprising: a generation unit. When the image of the region determined by the first determination unit to have the character attribute is a character image that is difficult to separate from the background by the second determination unit, the processing unit determines the character attribute. 2. The image processing apparatus according to claim 1, wherein the binary compression processing is not performed on the image of the area determined by the first determination unit to be. The second determination means determines whether the image of the area determined by the first determination means to have the character attribute is a character image that is easily separated from the background or a character image that is difficult to be separated from the background. The image processing apparatus according to claim 1, wherein the determination is performed based on the number of edges in the image of the area. Further comprising character recognition means for performing character recognition processing on the image of the area determined by the first determination means to have the character attribute,
4. The image processing apparatus according to claim 1, wherein the generation unit generates the file of the image including the character code obtained by the character recognition processing. The processing means performs the multi-value compression processing on the image of the area determined by the image attribute and the first determination means,
The generation unit determines whether the region is an image in which the one of the binary compression process and the multi-value compression process has been determined by the first determination unit as the character attribute, and the image attribute. 5. The file of the image is generated based on the image of the area in which the multi-value compression processing has been performed, which is determined by the first determination means to be present. The image processing device according to item 1. The image processing apparatus according to claim 1, further comprising a reading unit that reads a document to generate an image. A first determining step of determining whether the attribute of the area in the image is at least a character attribute or an image attribute based on the binarization result of the image ;
Wherein either said as a text attribute first determined at decision step area image is easy character image separated from the background, and a second determination step of determining whether the hard character image separated from the background ,
If it is determined in the second determination step that the image of the area determined in the first determination step as the character attribute is a character image that can be easily separated from the background , binary compression processing is performed on the image in the area. If the image of the area determined to have the character attribute in the first determination step is a character image that is difficult to separate from the background in the second determination step, the image of the area is displayed . A processing step for executing multi-value compression processing,
A file of the image is generated based on at least the image of the area that has been determined in the first determination step to have the character attribute and in which one of the binary compression processing and the multi-value compression processing has been executed. A method of controlling an image processing device, comprising: a generation step. A program for causing a computer of an image processing apparatus to execute the control method according to claim 7.