JPH04356873A - Adaptive encoding system for color document image - Google Patents

Abstract

PURPOSE:To clearly reproduce characters and to obtain high quality images having no density nonuniformity on the base by extracting a base area excepting for characters as well, encoding the characters and the base area as binary images and executing natural image encoding for remaining picture pattern areas. CONSTITUTION:The image data from a scanner 1 are stored in a buffer 2 and outputted to an area separation part 3 for each block of 8X8 picture elements, and characters, base block and picture pattern are judged. According to the judged result of the separation part 3, a switch 4 is operated to supply the characters and the image data in the base block to a binary image encoder part 5 and to supply the picture pattern block to a natural image encoder part 6. At the encoder part 5, the image data are binarized by a prescribed threshold value and encoded to binary images. At the encoder part 6, encoding for natural image is executed while defining the blocks excepting for the characters and the base block as picture pattern areas. Thus, the characters are made clear, and high quality images can be obtained without density nonuniformity on the base.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to an adaptive encoding method for color document images containing a mixture of text and pictures, and is applicable to various input/output devices for color document images such as electronic filing systems, color facsimiles, and color copies. It is suitable for this purpose.

0002

2. Description of the Related Art In recent years, research into encoding systems for color document images containing a mixture of text and pictures has become active. Such an encoding method is required because color document images require a larger amount of data than conventional monochrome document images, so when building a database by importing image data from an image input device such as a scanner, This is because efficient compression processing to reduce the amount of data is essential. In addition, as a natural image encoding method, ADCT (Adaptive Disc
This is because, if the createCosineTransform) method is adopted as a standardized method, mosquito noise will occur in edge portions such as characters, and image quality deterioration will be noticeable. Reversible entropy coding is used for images where resolution is important, such as text, and irreversible, multi-valued encoding methods such as band compression are used for images where gradation is important, such as natural images. is more suitable.

[0003] Therefore, edge components of characters are extracted from the original image, the edge components are encoded using dynamic arithmetic codes, and ADCT is used for the remaining components after removing the edge components from the original image. ``One encoding method for mixed text/image documents'' (Proceedings of the 1990 National Conference of the Institute of Image Electronics Engineers, pp. 131-136) has been proposed.

[0004] In this method, edge components of characters are first extracted, and the components are binarized to generate a character pattern, which is then arithmetic encoded. Next, the density of the extracted character pattern portion is replaced with the average density of the surrounding background portion to generate an image in which edge components are removed from the original image, and this image is encoded by ADCT.

The ADCT method, which separates edge components and adaptively encodes them in this way, is an image compression method for natural images, as described above, and is basically a method that cuts high-frequency components. Therefore, when applying to images such as text, the image quality around the edges will deteriorate unless the compression rate is lowered to reduce the amount of high-frequency components cut. If the edge components are separated, then binarized and reversibly encoded,
Since ADCT is applied to the image from which the high-frequency components of the edge portion have been removed, the image quality can be improved even with the same amount of code as a whole.

[0006]

By the way, in the conventional method described above, the background portion of the paper surface on which the color document image is drawn is regarded as part of the natural image and the ADCT method is applied. However, when considering a blank original used for color document images, the background part does not contain any intentional information, and the density unevenness when read with an image scanner is due to the quality of the paper used. There is.

Therefore, for such a background part, rather than faithfully reproducing the gradation at the time of input, it is better to replace it with a constant density to eliminate density unevenness and improve image quality. Furthermore, in terms of encoding efficiency, it is much more efficient to replace the background with a constant density value than to encode it by regarding the background as a natural image. Therefore, in this invention, by extracting the background part in addition to the characters, compared to the conventional method described above,
The purpose is to perform high compression and high quality encoding.

[0008] Further, the present invention focuses on the fact that natural images often do not require such high resolution, and it is possible to perform highly compressed encoding by lowering the resolution of the picture area and encoding it. purpose. Furthermore, this invention
We focused on the fact that the characteristics of the natural image area and the background area are significantly different from each other in the picture area from which the character pattern has been removed from the original image, and by using different encoding methods for the natural image area and the background area, we achieved high compression. The purpose is to perform encoding.

[0009]

[Means for Solving the Problems] The adaptive encoding method for color document images according to the present invention processes color document images in which natural images with gradation and characters are mixed in blocks of M×N pixels. It is determined whether the block is a background area or a picture area, and based on this determination result, the image data of the block determined to be a text/texture area is encoded in a manner suitable for a binary image, and the image data of the block determined to be a picture area is encoded. The image data is characterized by being encoded in a way that is suitable for the picture area.

0010

[Operation] In the adaptive encoding system for color document images according to the present invention, background regions are also extracted in addition to characters, the characters and background regions are encoded as binary images, and the remaining picture regions are encoded as natural images. Therefore, compared to the conventional method in which the background area is used as a picture area, characters are reproduced more clearly and the background has even density unevenness, resulting in a high-quality image.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing an embodiment of an adaptive encoding method for color document images according to the present invention. Figure 1
, the scanner 1 reads the target color document image, and because of the need to read characters, for example, 4
It has a resolution of about 00 dpi. The image data output from this scanner 1 is three types of image data of three primary colors RGB, the data value is 256 gradations for each color, and the white pixel is [
R, G, B] = [0, 0, 0], the black pixel is [R, G, B
] = [ 256, 256, 256 ].

Image data output from the scanner 1 is output in the order of raster scanning and is temporarily stored in a buffer 2 that stores several lines of image data. buffer 2
The image data stored in the image area separation unit 3 is processed block by block.
It is determined whether the block is a character/background block or a pattern (photo, halftone dot, solid, etc.) block. One block is 8x8 pixels or 16x according to the ADCT method.
It consists of 16 pixels. The configuration and determination operation of the image area separation unit 3 will be described later.

[0013] The characters referred to here are black characters that are comparable to printed characters, and pale characters such as those written in pencil or colored characters are images. Furthermore, the background refers to the background of paper as white as copy paper, and does not include background with low brightness such as newspaper. Also,
Highlights of photographs and halftone dots are treated as the image and are not included in the background.

The switch section 4 supplies the image data to the binary image encoding section 5 if the block is a character/background block based on the determination result of the image area separation section 3, and if the block is a picture block. If so, it is supplied to the natural image encoding unit 6. Therefore, since natural image encoding is performed only on picture areas, the compression rate can be significantly increased for images with few picture areas.

The binary image encoding unit 5 performs binary image encoding by binarizing the supplied image data of the character/background block using a predetermined threshold. Binarization is performed on edge-enhanced data in order to prevent the thickness of characters from changing depending on the threshold value. Furthermore, since the binary region is given in units of blocks, each block is raster-scanned and converted into a one-dimensional data series, and then entropy encoded using Huffman code, arithmetic code, or the like. Note that the details of the scanning method will be described later.

The natural image encoding unit 6 performs encoding for natural images using blocks other than character/background blocks as picture areas. Since the picture area also includes a halftone dot area, smoothing is performed to remove the shape of the halftone dots, and then orthogonal transform encoding such as ADCT is performed. The reason for removing the shape of halftone dots in advance is that halftone dots are originally removed to improve the apparent gradation with a small number of gradations (however, there are exceptions to some halftone dots with a low number of lines). This is because there is no need to leave the shape of the halftone dots intact, and ADCT has very low compression efficiency for halftone images.

In the area information encoding unit 7, the image area separation circuit 3
The information regarding the area determined in is encoded as area information. This area information is block-based information, and is binary information indicating whether it is a character/background area or another area.
The amount of data is quite small. Even if it simply has the same autocorrelation as characters and background, the amount of data is about 1/25
It becomes 6. Since the area information is also binary, entropy encoding is performed in the same way as for character/background areas.

Each data encoded in each of the encoding sections 5 to 7 is converted by a formatting section 8 into a format suitable for the application, and then stored in an external storage device 9.

Next, the image area separation section 3 will be explained with reference to the block diagram shown in FIG. The image area separation unit 3 converts the image data for each block supplied from the buffer 2 into MTF (M
oduration Transfer Function
on) is supplied to the correction section 10 and the white background extraction section 11. M
The TF correction unit 10 performs MTF correction on the image data, converts it into a binary value, and outputs the image data as resolution-oriented data to the edge extraction unit 12.
Output to.

The white background extracting unit 11 extracts a white background area by determining whether or not each block of image data supplied from the buffer 2 is a white background on a pixel by pixel basis. Although density information is simply used to determine whether or not the background is white pixel by pixel, many non-white background pixels remain in the background area due to density unevenness in the background area itself, so expansion processing is performed in units of blocks. Since non-white background images in the background appear to occur in a random pattern, the determination is made based on the number of white background pixels in the block. For example, when the block size is 16×16 pixels, if approximately 80% of the pixels are white, the entire block is treated as a white background block. However, in order to avoid misjudging a photo area as a white background, if there is even one color pixel in the block, it is treated as a picture block. However, in these processes, the white parts in the photograph and the highlighted parts in the halftone dots are still determined to be white background, so halftone dot detection is performed to prevent the halftone dot area from being mistakenly determined as the background. If there is an area determined to be a halftone dot in a white background block, that block is changed to a pattern block. A white background area is extracted by these processes.

The edge extraction section 12 separates edges of characters. Since there are many erroneous separations when edge extraction is performed on a pixel-by-pixel basis, the determination is basically made on a block-by-block basis. Although a method of simply determining based on the number of edge pixels may be considered as this determination method, in order to improve extraction accuracy, extraction is performed based on the shape pattern of edge pixels. The reason for this is that character edges generally have continuity. The shape pattern is vertical,
Prepare patterns that are continuous in the horizontal and diagonal directions depending on the size of the block. Recently, DTP (Desktop/
Since it is difficult to accurately separate the solid black area inside large characters, which is often found in printing (publishing), from the solid black area, it is treated as a picture area.

The comprehensive determination section 13 determines that the obtained edge area and white background area together are a character/background area based on the determination results of the white background extraction section 11 and the edge extraction section 12. However, since encoding is performed in units of 16×16 pixel blocks, character/background regions are determined in units of blocks of this size. If there is an edge or a white background block, it is determined that it is a character/background area. However, if a color pixel exists in an area that is neither a character nor a white background in a block, it is treated as a picture area. Finally, in order to change the white background block surrounded by the picture blocks in the photograph to a picture block, the picture blocks are expanded by one block in the vertical and horizontal directions.

FIG. 3 is a block diagram showing another embodiment of the image area separation section 3. As shown in FIG. In this example, buffer 2
The image data supplied from the CPU 11 is supplied to a black character extracting section 20 and a white pixel extracting section 21, respectively, to perform black character extraction and white pixel extraction. The black character extraction unit 20 extracts black characters as characters. However, it is difficult to accurately separate the solid black area inside large characters from the solid black area in the pattern, so
Treated as a picture area. As an algorithm for extracting black characters, for example, an image area separation method based on a "color image processing device" (Japanese Patent Application No. 2-214140) previously submitted by the present applicant is used. In this case, since the density level of the white background is expected to increase around the characters, the character area is expanded to include the surrounding background as the character area. This process prevents a picture area from occurring between the characters and the white background. The extraction result is stored in the buffer memory 22 in binary logic as "1" for the character area and "0" for the non-character area.

The white pixel extraction unit 21 treats each pixel as a non-white pixel if the density level of the G (green) component of the pixel is equal to or higher than a threshold Th, and as a white pixel if it is less than the threshold Th. In general, when the background density of the original is limited to the density of copy paper or less, that is, when the background of a newspaper or low-quality printed original is not detected as a background area, the density level of the highlight area in the picture (photo or halftone dot) is often higher than the density level of the background area of the original. Therefore, even if the determination is made on a pixel-by-pixel basis using a simple threshold value, there will be a difference in characteristics such that white pixels are densely distributed in the background region and white pixels are sparsely distributed in the highlight region. The extraction results are stored in the buffer memory 23 as logic "1" for white pixels and logic "0" for non-white pixels.

The extraction results of the black character extraction unit 20 and the white pixel extraction unit 21 stored in the buffer memories 22 and 23 are logically summed by an OR circuit 24 and stored in a buffer memory 25. If the result of the logical OR is logical "1" in pixel units, it becomes a character/background area, and if it is logical "0", it becomes a picture area.

By the way, in the ADCT method, which is a coding method for natural still images, generally 8×8 pixels or 16×1 pixels are used.
Since encoding is performed block by block with six pixels as one block, determination of character/background areas must also be determined using a block size suitable for such an encoding method. Therefore, the comprehensive determination unit 26 reads out the pixel data stored in the buffer memory 25 in units of, for example, 16×16 pixels, and determines whether the block is a text/background area or a picture area. The determination uses the number of black characters and white pixels in the block. As described above, white pixels are sparsely distributed in the highlight area, and white pixels are densely distributed in the background area. Therefore, in order to avoid misjudging highlight areas as text/background areas, we set the number of black characters and white pixels included in the block as a threshold (approximately 180 to 230 pixels), and set the number of black characters and white pixels included in the block to be greater than or equal to the threshold. Set the block as a character/text block. Finally, in order to prevent erroneous separation in the picture area, the text/background blocks surrounded by the picture blocks are changed to picture blocks. Through the above processing, character/background regions can be extracted.

Next, a scanning method for encoding image data stored in blocks in the buffer 2 will be described with reference to FIGS. 4(a) to 4(c). Figure 4 (a
) is a method in which pixel data is scanned and encoded block by block. In this method, character/background blocks are encoded while raster scanning within the block, and picture blocks are encoded using AD.
Perform block encoding such as CT. This method requires less memory capacity because it can process in blocks, but it cannot be implemented when reference pixels are required, such as in dynamic arithmetic codes.

The scanning method shown in FIG. 4(b) is a method of encoding in units of consecutive blocks of the same attribute. In this method, for character/background blocks, multiple blocks that are continuous in the row direction are collectively encoded in raster scan order, and for picture blocks, each block is encoded as in the case of Figure (a). Therefore, the number of blocks processed at one time is variable. When reference pixels are used in this method, such as in dynamic arithmetic codes, the pixels of the picture block use either a value of "0" or "1".

The scanning method shown in FIG. 4(c) is a method of encoding one block row by row. This method uses characters and
The background blocks are encoded in the raster scan order for each block row, and the pixels of the picture blocks are not jumped and encoded. As for the picture blocks, each block is encoded as in the case of Figure (a). When dynamic arithmetic codes are used in this method, the pixels of the picture block use either a value of ``0'' or ``1'', as in the case of Figure (b).

Next, with reference to the block diagram shown in FIG. 5, the binary image encoding section 5 when the scanning method shown in FIG. 4(c) is applied will be explained. In FIG. 5, image data supplied from the switch unit 4 (FIG. 1) is subjected to MTF correction in an MTF correction unit 30, and is compared with a predetermined threshold value in a binarization circuit 31 to be binarized.
is stored in

If the result of the judgment in the image area separation unit 3 is that the image data is character/background area data, the line buffer memory 32 writes the binarized data in the binarization circuit 31 to the corresponding address; If the data is in the picture area, a logic "1" is written. This write control is performed by the address controller 33.

When data for one block row is written to the line buffer memory 32, the address controller 3
3, the data of the pixel of interest and the reference pixel are read out from the memory 32, and are supplied to the encoding unit 34 and encoded. A template as shown in FIG. 6 is used to extract reference pixels. Further, as the encoding unit 34, JB
A QM-Coder encoding unit by IG is used.

To explain the line buffer memory 32 in more detail, this memory can store data for two block rows with one pixel and one bit, and is configured to enable pipeline operation. For example, one block line is 160 horizontally
In the case of 0 pixel and 16 pixels vertically, two sets of 1604×16 bit line buffer memories are required. The reason why there are 4 extra pixels horizontally is because there are 2 pixels on the left and right as a reserve area for calculating reference pixels. Letting these two sets of line buffer memories be Ba and Bb, respectively, the values of all buffer memories are set to "0" at the start of processing. Then, when the first write data is input, data is sequentially written in the raster direction from the position (2,0) of the buffer memory Bb. When all the data is written to the buffer memory Bb, the line buffer memory Ba is written.
The reference pixel and the target pixel are read from the buffer memory Bb. However, the values of the 15th and 16th lines of the buffer memory Ba are used as the reference pixels of the first and second lines and the line two lines before. For example, when (2,0) of buffer memory Bb is the pixel of interest, the position of the reference pixel is (0,0) of buffer memory Bb.
0), (1,0) and (0,15) of buffer memory Ba
, (1,15), (2,15), (3,15), (4,
15), (1, 14), (2, 14), (3, 14).

Next, another encoding method of the present invention will be explained with reference to FIG. This method is a method in which the resolution of the picture area is lowered and encoded. In other words, when reading a color document image with a scanner, the current technology is only able to read the entire surface of a document at the same resolution, so generally it is 400 dp so that small characters can be read.
A resolution higher than i is required. However, natural images often do not require such high resolution. However, until now, encoding has been performed for the picture area using the resolution at the time of input. This is because in the ADCT method used as a natural image encoding method, the minimum block size is 8 x 8 pixels or 16 x 16 pixels. This is because the block size has to be increased, which increases the buffer memory capacity.

Therefore, in this method, the resolution of the picture area is lowered and encoded without increasing the size of the judgment block for image area separation. In the example in Figure 7, the required resolution for the picture area is 100 dpi, and the required resolution for the text/background area is 40 dpi.
It is encoded as 0dpi. Therefore, the picture area is reduced to one quarter in both the vertical and horizontal directions. That is, assuming that 8×8 pixels are one block, the number of pixels to be encoded per block of the picture area is 2×2 pixels, a total of 4 pixels, and the amount of data is compressed to 1/16. The reduction process is simply performed by smoothing process. For encoding, reversible pixel-by-pixel encoding such as DPCM is used for the picture area. 2, such as dynamic arithmetic code, for characters and background areas.
Use value image encoding. Since both methods refer to peripheral pixels, they are performed for each block row.

In FIG. 7, a square portion surrounded by a solid line represents one block. The picture block is reduced to 4 pixels per block by smoothing, as shown by the broken line. The picture block is encoded using pixels a11, a12, b11.
, b12, ..., and when the rightmost pixel is encoded, the next line starts to be encoded, and pixels a21, a22, b2 are encoded.
1, b22, . . . are encoded in this order. When using surrounding pixels to predict the pixel value of interest, if the surrounding pixels belong to the text/background area, set the pixel value included in the text/background area to "0".
The pixel of interest is predicted and encoded. The character/background area is encoded pixel by pixel in the order of raster scan, as shown by the arrow in the figure. When reference pixels are used as an encoding method such as dynamic arithmetic encoding, the pixel value included in the picture area is assumed to be "0", and the pixel of interest is predicted and encoded.

According to this method, since the image in the picture area is encoded at an appropriate resolution, a higher compression rate is possible than when encoding the original image data as is.

Next, another embodiment of the natural image encoding section 6 will be described with reference to the block diagram shown in FIG. In this example, we focus on the fact that the natural image area and the background area have significantly different characteristics even in the picture area where the character pattern is removed from the input image, and use different encoding methods for the natural image area and the background area. In this way, higher encoding efficiency can be obtained. In other words, since the natural image area includes color changes and textures, as a result of orthogonal transformation, DC
A relatively large number of low frequency components other than (DC) components appear. On the other hand, in the background region, the density is approximately constant when viewed locally, so that in the spatial frequency region there is almost only a DC component.

Therefore, in this embodiment, the remaining image after removing the character pattern from the input image is separated into a natural image area and a background area, and the encoding method is adaptively changed according to each area. , and performs high-compression encoding. To change the encoding method, there are two methods: changing the encoding method itself and changing only the parameters. The compression ratio is improved by switching the tables.

In FIG. 8, the image area separation unit 40 converts the image data after removing the character pattern from the original image data into M×
It is determined whether each block of N pixels is a background area or a natural area. The density of white pixels is used for this determination. That is, if a block contains, for example, 90% or more of pixels whose luminance component density level is equal to or less than the threshold Th, that block is determined to be a background block. This result is sent to the area information encoding section 7 (FIG. 1) as the attribute information of the block, and is also supplied to the control terminal of the switch section 43 for selecting either the Huffman encoding table 41 or 42. The Huffman encoding table 41 is a table for natural images, and the Huffman encoding table 42 is a table for backgrounds.

In the ADCT encoding unit 44, as shown in FIG. 9, a discrete cosine transformation unit 50 performs two-dimensional discrete cosine transformation (DC
T) and transform it into the spatial frequency domain. As a result of the DCT calculation, the upper left of the block becomes a DC (direct current) component, and the closer to the lower right, the higher the frequency component becomes. In the case of a natural image, since the correlation between pixels is high, a large DCT coefficient often appears in the low frequency component at the upper left. Therefore, the quantization unit 51 performs compression by dividing the low frequency components of the DCT coefficients by a small value and dividing the high frequency components by a large value. The divisor for each frequency component at this time is given in advance to the quantization table 52. Since the quantization coefficient values in the high frequency component area after quantization continue to have many "0"s, they are scanned in zigzag order as shown in Figure 10 and converted into run legs of "0" and component values other than "0". , the Huffman encoding section 53 performs Huffman encoding on this while referring to the Huffman encoding table 54. This Huffman encoding table 54 is the Huffman encoding table 4 shown in FIG.
1 and 42.

Huffman encoding is a type of entropy encoding that assigns short codes to events with a high probability of occurrence. Huffman encoding is performed separately into a DC component and an AC component. In encoding the DC component, basically, the difference with the DC component value of the previous block is encoded instead of the DC component value itself. Assigning Huffman codes to all data,
Since a huge amount of storage space is required to store the table, the DC component value is decomposed into categories and additional bits. A category is a collection of component values, and is equal to the minimum number of bits C required to represent a component value. If the component value is positive, the additional bits are the lower C bits of that value, and if the component value is negative, the additional bits are the lower C bits of the value obtained by subtracting "1" from the component value. For example, the difference value of the DC component is “10” (1
In the case of a decimal number), the category is "4" and the additional bit is "0010". A Huffman code will be assigned to this category. FIG. 11 shows an example of a Huffman code table for a luminance signal.

The code word for the AC component consists of a run length of "0" and a category value of the AC component of non-"0". For example, if the quantization coefficients shown in FIG. 10 are obtained, the results of zigzag scan are (0,3), (
4,3), (10,1), EOB (End Of Bl
ock) is sent. The first number in parentheses indicates the "0" run length, and the subsequent numbers are the non-"0" AC component values. From this event, “NNNNSSSS
+ additional bits. "NNNN" indicates the run length. Since the run length is up to 15, it is represented by 4 bits. The next “SSSS” is a category of non-“0” values. Since this is also expressed using 4 bits, a code word for Huffman encoding is composed of a total of 8 bits. The additional bits are created in the same way as the DC component. “EOB” indicates that the following values of the block are all zero, and is represented by “0x00”. The Huffman encoding table for the AC component is
NNNNSSSS", compression is performed by assigning short code words to code words that appear frequently.

In natural images, the quantization coefficient of the luminance component is
Since the value "0" appears with a relatively high probability in the first and second components of the AC component, "0x01", "0x02", "
Short codewords of 2 to 4 bits are also assigned to the codewords ``0x03'' and ``0x11.'' However, in the background area, even the first and second components are non-zero.
The value of almost never appears. In other words, the probability of "EOB" appearing becomes extremely high. Therefore, in the background area “EOB
The compression rate can be greatly improved by assigning the code closest to "EOB". For this reason, it would be appropriate to assign a code of 1 or 2 bits to "EOB".

Returning to FIG. 8, in the ADCT encoding section 44,
As described above, the Huffman encoding tables 41 and 42 are adaptively switched according to the attributes of the block. Therefore, the DCT transformer 50 and the quantizer 5 other than the table
1. The Huffman encoding unit 53 and the like are common to all block data. The encoding result is combined with the outputs of other encoding sections and recorded in the external storage device 9 (FIG. 1).

According to this embodiment, a natural image area and a background area are separated from an image obtained by removing characters from the original image, and encoding is performed appropriately for each by changing the encoding parameters (encoding table). Therefore, encoding efficiency can be improved with a simple configuration.

The decoding process of the image data thus stored in the external storage device 9 is performed by the decoding device shown in FIG. Decoding is the inverse of encoding. However, it is not necessary to perform image area separation. First, data is read from the external storage device 9 to the deformatting unit 60 and deformatted, the binary image code is sent to the binary image decoding unit 61, the natural image code is sent to the natural image decoding unit 62, and the area information is sent to the area Each is output to the information decoding section 63. Decoding of binary image codes and natural image codes in the binary image decoding unit 61 and the natural image decoding unit 62 is performed by the area information decoding unit 63.
controlled according to the area information restored in . Then, the final decoding is performed by the general decoding unit 64 in a form suitable for the application request. For example, printer 6
5, the images are restored in raster order. Furthermore, when displaying the entire image on the display 66, a reduced image that has been thinned out is played back.

[0048]

According to the present invention, since the background area is extracted in addition to the characters, the characters and the white area can be encoded as a binary image, and only the remaining picture area needs to be encoded as a natural image. The compression rate can be significantly improved for documents with small picture areas. Furthermore, since the text/background area is converted into a binary image, the text is clearly reproduced, and the background is also free from density unevenness, making it possible to obtain a high-quality image.

[0049] Furthermore, since the background area is extracted by combining the results of white background detection and halftone dot detection in units of blocks, it is possible to accurately determine the background area in printed manuscripts where the image is expressed by halftone dots. . Furthermore, since the picture area is expanded in the end, if the periphery of the character/background block is surrounded by the picture area,
By changing the background block to a picture block, it is possible to reduce erroneous determination of picture areas and obtain a restored image of higher quality.

Furthermore, after obtaining black characters and white pixels individually in pixel units, image area separation is performed in block units based on the density of black characters or white pixels included in the block. - The background area can be extracted. Further, according to the present invention, since character/background blocks are encoded in units of blocks, encoding can be performed using a buffer memory with a small storage capacity. Furthermore, since encoding is performed in units of a plurality of consecutive blocks or in units of one block row, efficient encoding can be performed.

Furthermore, according to the present invention, since the natural image is encoded at an appropriate resolution, higher compression is possible than when encoding the original image as it is. Also,
According to this invention, the background area and the natural image area are separated from the image from which characters have been removed from the original image, and encoding is performed suitable for each by changing the encoding parameters. It is possible to perform conversion.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of an adaptive encoding method according to the present invention.

FIG. 2 is a block diagram showing one embodiment of the image area separation section in FIG. 1;

FIG. 3 is a block diagram showing another embodiment of the image area separation unit.

FIG. 4 is a diagram for explaining a scanning method when encoding image data in units of blocks.

FIG. 5 is a block diagram of a binary image encoding section.

FIG. 6 is a diagram showing a template for extracting reference pixels.

FIG. 7 is a diagram showing a method of encoding a picture area by lowering its resolution.

FIG. 8 is a diagram showing another embodiment of the natural image encoding section.

FIG. 9 is a block diagram of an ADCT encoding section in FIG. 8;

FIG. 10 is a diagram showing a zigzag scan of quantization coefficient values.

FIG. 11 is a diagram showing an example of a Huffman code table for a luminance signal.

FIG. 12 is a block diagram showing an example of a decoding device that decodes encoded image data.

[Explanation of symbols]

1 Scanner 2 Buffer 3 Image area separation section 5 Binary image encoding unit 6 Natural image encoding unit 7 Area information encoding unit 8 Format section 9 External storage device 10 MTF correction section 11 White background extraction part 12 Edge extraction part 13, 26 Comprehensive Judgment Department 20 Black character extraction part 21 White pixel extraction section

Claims (11)

[Claims] Claim 1: A color document image in which a natural image with gradation and text are mixed is determined in block units of M×N pixels as to whether it is a text/background area or a picture area, and the determination result is determined. The image data of the block determined to be the character/background area based on the above is subjected to encoding suitable for a binary image, and the image data of the block determined to be the above picture area is subjected to encoding suitable for the picture area. An adaptive encoding method for color document images. 2. In claim 1, the determination of the background area is performed by determining whether or not each pixel is a white pixel to obtain the number of white pixels for each block, and determining whether the number of white pixels in the block is a predetermined number. Adaptation to color document images characterized in that if the block is equal to or greater than a threshold value and does not contain color pixels, the block is treated as a white background block, and if there is no halftone area in this white background block, the block is treated as a background area. Encoding method. 3. In claim 1, the image area separation between the character/background area and the picture area is performed when an edge area or a background area exists in each block after extracting the character area and the background area. An adaptive encoding method for color document images characterized in that, if there are no color pixels, the block is determined to be a character/background area, and then the picture area is expanded. 4. In claim 1, the image area separation between the character/background area and the picture area includes extracting pixels around the character, extracting white pixels, and taking a logical sum of these extraction results. Extract characters or background pixels by
A color document image characterized in that when character/background areas are determined in units of M×N pixel blocks, if the density of characters or white pixels in the block is high, the block is determined to be a character/background block. adaptive coding scheme. 5. The adaptive encoding method for color document images according to claim 1, wherein the encoding of the character/background area is performed in units of the blocks. 6. The adaptive encoding method for color document images according to claim 1, wherein the character/background area is encoded in units of a plurality of character/background blocks that are continuous in the row direction. 7. The adaptive encoding method for color document images according to claim 1, wherein the character/background area is encoded in units of character/background blocks of one block row. 8. In claim 6 or 7, when the character/background block is encoded by a method that refers to surrounding pixels, the pixels included in the picture block are encoded by regarding them as a predetermined constant value. An adaptive encoding method for color document images. 9. The adaptive encoding method for color document images according to claim 1, wherein the picture area is encoded after being converted to a resolution lower than that of the character/background area. 10. Separating a text/background area and a picture area from a color document image in which a natural image with gradation and text coexist, further separating a background area from the picture area, and separating the text/background area from the picture area. It is characterized by performing encoding suitable for a binary image for the background area and performing encoding suitable for the pattern area using different parameters for the background area and the pattern area from which the background area has been removed. An adaptive encoding method for color document images. 11. In claim 10, the encoding suitable for the picture area is ADCT encoding, and the different parameters are a Huffman encoding table suitable for a background area and a Huffman encoding table suitable for a natural image. An adaptive encoding method for color document images characterized by the following.