JPH0998294A - Image information coder and decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To avoid deterioration in a reproduced image even when part of a background is segmented as a character or the like in mistake by providing a means conducting labelling based on density information and a means coding attribute data representing in which region the information is included. SOLUTION: A characteristic background separate section 21 separates an original image into a character part and a background part. The separated background is coded by a background coding section 22. The separated character part is fed to a skeleton transient separate section 23, in which a skeleton area and a transient area are separated based on a degree of change in density of surrounding picture elements. Data of the skeleton area are fed to a labelling section 26, connected and loabelled based on density information. The labelled skeleton area is reduced into 2-dimension by a skeleton area reduction section 27 and plural skeleton areas connected in longitudinal and lateral directions are separated. An area attribute discrimination section 28 discriminates which attribute the area has to generate and code the attribute data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention encodes image information,
The present invention relates to a decoding device, and more particularly to a device which encodes and decodes an image in which characters and photographs are mixed together with high reproducibility.

[0002]

2. Description of the Related Art As an image coding method, for example, I
S0 (international0rganizat
Ion for standardization) and CCITT (Commit consultatif)
lternatina ITelegraphie
At the joint of et Telephonique, JPEG (Jo) has been studied for international standardization.
int Photographic Coding E
xperts Group) encoding method, and JBI
G (Joint Bi-level image Co)
ding Experts Group) method.

In the JPEG system, the input image is subjected to discrete cosine transform, the transform coefficient is quantized, and the quantized index is entropy coded. When the input image is a natural image, the power is concentrated on the transform coefficient with a low spatial frequency, so the low-frequency transform coefficient is finely quantized, and the high-frequency transform coefficient is roughly quantized to reduce the data amount. can do. However, when the input image is an image containing many sharp edges such as characters and lines, it is difficult to reduce the amount of data because it is difficult to concentrate the power on the conversion coefficient having a low spatial frequency. If the high-frequency transform coefficient is roughly quantized roughly, the image quality of the decoded image is significantly deteriorated.

In order to solve the above problems, the present inventors have already proposed a method for reliably cutting out a character from the background portion and then separately encoding the character and the background to solve the above problems. However, the applicant has applied for it.

An image processing apparatus using this method, as shown in FIG. 6, extracts a binarization unit 2 for binarizing an input image 1 using a predetermined threshold value and an edge portion of the input image 1. An edge extraction unit 3, a character / background separation unit 4 that separates a character region from a natural image region by using edges for a binarized image, and a region area measurement unit 10 that measures the number of pixels of the separated region. , A background area removal unit 6 for removing the area having a large area and an area having a small area by threshold processing, a character area distribution measuring unit 7 for measuring the distribution of the character area, and a center of gravity of the area. An area centroid measuring unit 11 for measuring and a background deleting unit 8 by a peripheral distribution for deleting a background by using a distribution of character areas are provided.

In the binarization unit 2, the input image 1
Is binarized using a predetermined threshold, and character region candidates including the original natural image region are extracted. At the same time, the edge extraction unit 3 extracts the edge A from the input image 1 shown in FIG. In the character background separation unit 4, as shown in FIG.
By deleting the pixels in the vicinity of, the area extracted by the binarization unit is separated into the character area B and the background area C, and the separation result 5 is output. The area of the separated area is measured in the area area measuring unit 10. By the background deletion unit 6 depending on the area area, a predetermined darkness value th
If l and th2 (where thl> th2) are used and the area area is larger than thl or the area area is smaller than th2, the area is deleted. The character area distribution measuring unit 7 measures the cumulative number of pixels in the character area along a predetermined direction such as the vertical direction and the horizontal direction, that is, the peripheral distribution, and the result is shown in FIG. As described above, the range in which the character string exists, that is, the range D of the character string is detected. The area center of gravity measurement unit 11 measures the center of gravity of each of the extracted areas. In the background deletion unit 8 based on the marginal distribution, the center of gravity of each area is the range D of the character string.
If the center of gravity of each area is not within the range D of the character string, the area is determined to be the natural image area E, as shown in FIG.
As shown in (b), the natural image area E is deleted. Finally, the extracted character image 12 is output.

However, even if the accuracy of character extraction is improved from various viewpoints as described above, it is difficult to completely prevent a part of the background from being recognized as a part of the character. Then, even a slight misrecognition may greatly affect the reproducibility of the image. For example, in FIG.
As shown in FIG. 9B, the surrounding background portion is erroneously cut out as a character portion with respect to the original image (white portion is a character area) as shown in FIG. 9A, and the erroneously cut out portion is If it is combined with the authentic character part, confusion will occur between the two. That is, in the method of separately encoding the character area and the background area, attribute data indicating whether the target area is the character portion or the background portion is formed. At this time, if the regions having the same attribute are combined, it becomes impossible to distinguish the two regions, and one density information is assigned to both regions. Therefore, in the example of FIG. 9, the character area cannot be distinguished from the background area as shown in (c), and is reproduced with the same density.

The present inventors have already proposed a method of constructing attribute data (“Adaptive method selection and efficient encoding of character image based on three-valued MMR”, 1992 image code). Symposium (PCSJ9
2, pp235-238).

In a specific example, the original image is divided into blocks, for example, 2 × 2 blocks, and the attribute is determined for each block. The flat portion, the quasi-flat portion, and the edge portion are used as the attributes. The attribute determination criterion is, for example, the difference between the highest density and the lowest density in the target block, and the difference is a predetermined threshold (TH
If it is smaller than 1), it is determined to be a flat portion. The difference is TH
If it is greater than 1, the difference is set to another threshold value (TH2>
TH1), if it is smaller than the threshold value TH2, it is determined to be a quasi-flat portion, and if equal or larger, it is determined to be an edge portion.
For the block in the flat part, one-level BTC (Bloc
k Trans Coding) code, a quasi-flat part block is encoded using a 2-level BTC code, and an edge part block is encoded using a prediction code.

The distribution of various attribute regions, that is, attribute data, is a three-valued MMR (transition of attribute is represented by 1, 0, M
Encode with MR. MMR is a Modified Mod
It is coded by "ifed Read code". The attribute distribution is used when encoding and decoding each block.

[0011]

SUMMARY OF THE INVENTION The present invention has been made in consideration of the above circumstances, and cuts out components such as characters from an image such as a photograph in which components such as characters are superimposed,
Provided is an image information encoding device and a decoding device in which a reproduced image is not significantly deteriorated even when a part of the background is erroneously cut out as a character or the like when encoding the character and the background separately. Is intended.

[0012]

According to the present invention, in order to achieve the above object, a composite image formed by superimposing on a background image image elements such as characters whose density changes rapidly with respect to the surroundings. Is the first attribute area corresponding to the background image,
Means for dividing into a second attribute area corresponding to a skeleton area in the image element such as the character and a third attribute area corresponding to a peripheral transient area in the image element such as the character, and the second attribute area, The means for labeling based on information, the means for reducing the labeled second attribute area, and the respective areas of the image obtained by reducing the second attribute area are the first
Means for encoding attribute data representing which of the attribute area, the second attribute area and the third attribute area is included,
Means for determining a single density information for each of the reduced second attribute areas, which is defined by the attribute data, and encoding the density information, and a compression code with low reproducibility of high frequency components. In the image information coding apparatus, means for coding the first attribute area by a coding method and means for coding the third attribute area by a compression coding method with high reproducibility of high frequency components are provided. I have to.

Further, according to the present invention, means for decoding the encoded attribute data, means for modifying the attribute data so as to enlarge the second attribute area specified by the decoded attribute data, and code Means for decoding the converted density information for each of the second attribute areas, and means for reproducing pixel information of the second attribute area based on the corrected attribute data and the decoded density information, Means for decoding the encoded first attribute area to reproduce the pixel information of the first attribute area, and decoding the encoded third attribute area to reproduce the pixel information of the third attribute area. Means, and reproduced pixel information of the first attribute area,
The image information decoding apparatus is provided with means for synthesizing the reproduced pixel information of the second attribute area and the reproduced pixel area of the third attribute area.

According to the above configuration, the second attribute region corresponding to the skeleton region is reduced and, for example, a plurality of second attribute regions are connected.
Even if there is an attribute area, it can be separated by reduction operation,
Correct density information can be assigned to each, and deterioration on the reproduced image can be suppressed.

In this structure, if the width of the second attribute region is reduced and the line segment disappears, it is important not to perform the reduction. Therefore, a configuration may be adopted in which the reduction is stopped when the line segment of the second attribute region has the minimum line width.

Further, according to the present invention, a composite image formed by superimposing an image element such as a character whose density changes abruptly with respect to the surroundings on the background image is defined as a first attribute area corresponding to the background image
And a unit that divides the second attribute region corresponding to the region of the image element such as the character, a unit that labels the second attribute region based on the density information, and a unit that labels the second attribute region. A means for reducing, a means for encoding attribute data representing which of the first attribute area and the second attribute area each area of the image obtained by reducing the second attribute area is included, For each group of the reduced second attribute areas defined by the attribute data,
The image information coding apparatus is provided with means for determining a single piece of density information and coding the density information and means for coding the first attribute area.

According to this structure, even when there is no peripheral transient area, it is possible to code the image information in the same manner in which a character segmentation error is dealt with.

According to the present invention, means for decoding the encoded attribute data, and means for modifying the attribute data so as to enlarge the second attribute area specified by the decoded attribute data, Based on the means for decoding the encoded density information for each of the second attribute areas, and the corrected attribute data and the decoded density information,
Means for reproducing the pixel information of the second attribute area, means for decoding the encoded first attribute area to reproduce the pixel information of the first attribute area, and means for reproducing the pixel information of the first attribute area reproduced. The image information decoding apparatus is provided with means for synthesizing the pixel information and the reproduced pixel information of the second attribute area.

According to this structure, even when there is no peripheral transient area, it is possible to decode the image information in the same manner in which a character segmentation error is dealt with.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows an image information encoding apparatus according to an embodiment of the present invention. In this figure, the image information encoding apparatus includes a character / background separating unit 21, a background encoding unit 22, and a skeleton / transient separating unit 23. The transition area pixel coding unit 24, the skeleton region density coding unit 25, the skeleton region labeling unit 26, the skeleton region reduction unit 27, the region attribute determination unit 28, the attribute data coding unit 29, and the code data transmission unit 30.

First, the character / background separating unit 21 separates the original image into a character portion and a background portion. The original image is, for example, a photo image in which characters having a constant density are typeset. For example, black, white, or gray is used as the character. For black and white characters, a transitional region with rounded edges is provided around the skeleton region with a constant density to suppress the jaggedness of the character outline. Only the skeletal region is shown for gray. The character / image separation unit 21 can employ the conventional configuration based on the edge detection shown in FIG. 6, for example. Of course, it is possible to adopt various configurations such as dividing an image into blocks, and determining whether or not it is a character area based on the maximum density and the minimum density of each block.

The separated background portion is encoded by the background encoding portion 22. This encoding is, for example, similar to JPEG, D
It can be configured by a CT (discrete cosine transform) conversion unit 221, a transform coefficient quantization unit 222, and an entropy coding unit 223.

The separated character part is the skeleton / transient separation part 23.
The skeleton / transient separation unit 23 separates the skeleton region and the transient region based on the degree of change in the density of the surrounding pixels.
The pixel data of the transient area is sent to the transient area pixel coding unit 24, and is coded in pixel units, for example, by the Huffman code.

The data of the skeleton region is sent to the labeling unit 26, where it is connected and labeled based on the density information. The representative density of the connected area is determined based on, for example, the average value and the intermediate value of the density of the area, and the density value is assigned to each label. This density value is coded by the skeleton area density coding unit 25 using, for example, a Huffman code.

The labeled skeleton region (connecting region) is two-dimensionally reduced by the skeleton region reducing unit 27. For example, the skeleton region is scanned in the vertical direction from top to bottom, and the first one pixel of the skeleton region is incorporated into the transient region. Similarly, the first one pixel of the skeleton region is incorporated into the transient region when scanning from left to right in the horizontal direction. The skeletal region reducing unit 27 can separate a plurality of skeletal regions that are combined in the vertical and horizontal directions.
This will be described later in detail with reference to FIG.

When the line width of the character is narrow and the width of the skeleton region is 1 pixel, if the reduction processing is performed in the same manner as described above, a part of the skeleton region may be interrupted or the whole skeleton may disappear. There is a risk. In order to prevent this, in this embodiment, reduction is prohibited when the line width is 1 pixel.

The area attribute discriminating unit 28 discriminates which attributes (background area, skeleton area and transition area) the area has, and generates attribute data. This attribute data is encoded by the attribute data encoder 29. The attribute data is
Since it consists of three attributes, it cannot be encoded with a binary MMR code. Therefore, in the case of ternary value, the same value is not obtained after the change. Therefore, a rule for assigning 1 or 0 is provided according to which transition is made at the change point, and three attributes are binaryly MMR encoded. I have to. This is a three-valued MMR
Called encoding.

Background code section 22, transition area pixel code section 2
4, skeleton region density encoding unit 25 and attribute data encoding unit 2
The data coded in 9 is combined by the coded data sending unit 30 and sent through the transmission line.

FIG. 2 shows an image information decoding apparatus for decoding the image information encoded by the image information encoding apparatus of FIG. 1. In this figure, the image information decoding apparatus is an encoded data separating section 31. , Background decoding unit 32, attribute data decoding unit 3
3, a skeleton region density decoding unit 34, a skeleton region expansion unit 35, a transient region pixel decoding unit 36, and an image synthesizing unit 37. The encoded data is first separated by the encoded data separating unit 31 into background data, attribute data, skeleton region density data and transient region pixel data. The background data is decoded by the background decoding unit 32. The decoding unit 32 includes, for example, an entropy decoding unit 321, an inverse quantization unit 322, and an inverse DCT conversion unit 3.
23. The attribute data is the attribute data decoding unit 33.
In the three-valued MMR decoding (which is the inverse conversion of the previous three-valued MMR.
The attribute is decoded by the above rule and MMR decoding).
The decoded data is enlarged by the skeleton region enlargement unit 35 in the vertical and horizontal directions, for example, one pixel at a time. This enlargement processing is the reverse of the reduction processing executed by the skeleton area reduction unit 27 in FIG. The skeleton area density data is subjected to, for example, MMR decoding by the skeleton area density data decoding unit 34. The transient area pixel data is converted by the transient area pixel decoding unit 36 into
For example, Huffman decoding is performed. The image synthesizing unit 37 synthesizes images based on the decoded data.

FIG. 3 illustrates the mode of encoding and decoding using the image information encoding apparatus of FIG. 1 and the image information decoding apparatus of FIG. 2 described above. The original image shown is separated by the background / character separation unit 21. In this example, the background portion, which happens to be adjacent to the skeleton region (white region) of the character portion, has the character of a character, and is therefore mistakenly detected as a skeleton region by edge detection or the like (b).
In addition, in (b), a black portion indicates a portion cut out as a skeleton region (character portion). The skeletal region and the background had in this case different densities and are therefore labeled separately (C). After that, when the skeleton region is reduced, the adjacent skeleton region is separated (d). After this, (e)
As shown in, the image is divided into a skeleton region (shown in black), a transient region (shown in gray) and a background region (shown in white) to generate attribute data. Although encoding is performed by this attribute data, it is possible to prevent the skeleton region and the background region which are erroneously combined from being divided, and different density information to be added, thereby sending wrong density information.

In the decoding process, the skeleton region is enlarged and returned to the original state, and as a result, the restored image becomes faithful to the original image of (a) (f).

FIG. 4 shows the original image, the reproduced image in this embodiment, and the reproduced image in the JPEG system in comparison with each other. According to this embodiment, the fluctuation around the characters found in the JPEG system is shown. Is significantly suppressed and the image quality is improved. Further, FIG. 5 shows a difference image between the original image and the image of this embodiment, and a similar original image and JP in the example of FIG.
It is shown by comparing the difference image with the image of the EG method. It can be seen that in this embodiment, there is almost no difference from the original image.

In the above embodiment, the two attributes consisting of the skeleton region and the transient region are assigned to the character region, but only the attributes of the skeleton region may be assigned. In this case, it is possible to encode with a code such as a normal MMR instead of the ternary MMR. Even in this case, even if the skeleton region and the surrounding background region are connected and recognized as a skeleton region by the reduction and enlargement processing of the skeleton region, the correct density information can be similarly added.

[0035]

As described above, according to the present invention, even if a part of the background area is erroneously cut out as a character area, the density information can be correctly coded, resulting in a character cutting error. It is possible to suppress deterioration of image quality.

[Brief description of drawings]

FIG. 1 is a block diagram showing an image information encoding device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing an image information decoding device of the above-described embodiment.

FIG. 3 is a diagram for explaining the operation of the above embodiment.

FIG. 4 is a diagram comparing a reproduced image of the above-described embodiment with an original image and a reproduced image obtained by a conventional method.

FIG. 5 shows the difference between the reproduced image and the original image in the above-described embodiment,
It is a figure which compares with the reproduction | regeneration image and the original image of a conventional system, and shows it.

FIG. 6 is a block diagram showing a conventional example.

FIG. 7 is a diagram illustrating an operation of a conventional example.

FIG. 8 is a diagram illustrating an operation of a conventional example.

FIG. 9 is a diagram illustrating deterioration of image quality due to a character cutting error in a conventional example.

[Explanation of symbols]

 21 Character / Background Separation Unit 22 Background Code Unit 23 Skeleton / Transient Separation Unit 24 Transient Region Pixel Code Unit 25 Skeleton Region Density Code Unit 26 Skeleton Region Labeling Unit 27 Skeleton Region Reduction Unit 28 Region Attribute Discrimination Unit 29 Attribute Data Encoding Unit 30 Coded data transmission unit 31 Coded data separation unit 32 Background decoding unit 33 Attribute data decoding unit 34 Skeleton region density decoding unit 35 Skeleton region expansion unit 36 Transient region pixel decoding unit 37 Image combining unit

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] December 21, 1995

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] Name of invention

[Correction method] Change

[Correction contents]

Title: Image information encoding device and decoding device

Claims (5)

[Claims] 1. A composite image formed by superimposing an image element, such as characters, whose density changes rapidly with respect to the surroundings on a background image,
A first attribute area corresponding to the background image, a second attribute area corresponding to a skeleton area in the image element such as the character, and a third attribute area corresponding to a peripheral area in the image element such as the character. , Means for labeling the second attribute area based on the density information, means for reducing the labeled second attribute area, and each area of the image obtained by reducing the second attribute area A means for encoding attribute data representing which of the first attribute area, the second attribute area and the third attribute area is included, and a group of the reduced second data defined by the attribute data. A unit for determining a single density information for each attribute region and encoding the density information; a unit for encoding the first attribute region by a compression encoding method with low reproducibility of high frequency components; High reproducibility of ingredients In condensation coding method, the image information encoding apparatus, characterized in that it comprises a means for encoding said third attribute area. 2. An image information decoding apparatus for decoding image information coded by the image information coding apparatus according to claim 1, wherein: means for decoding the coded attribute data; Means for modifying the attribute data so as to expand the second attribute area specified by the attribute data; means for decoding encoded density information for each of the second attribute areas; and the modified attribute data and Means for reproducing pixel information of the second attribute area based on the decoded density information; means for decoding the encoded first attribute area to reproduce pixel information of the first attribute area Means for decoding the coded third attribute area to reproduce pixel information of the third attribute area, reproduced pixel information of the first attribute area, and reproduced second attribute area of the second attribute area. Pixel Distribution and the image information decoding apparatus characterized by comprising a means for combining the pixel region of the third attribute area is the reproduction. 3. The image information coding apparatus according to claim 1, wherein the reduction of the second attribute area is stopped when the line segment of the second attribute area reaches the minimum line width. 4. A composite image formed by superimposing an image element, such as characters, whose density changes rapidly with respect to the surroundings on a background image,
Means for dividing into a first attribute area corresponding to the background image and a second attribute area corresponding to the area of the image element such as the character, and means for labeling the second attribute area based on the density information; A means for reducing the labeled second attribute area, and an attribute indicating which area of the first attribute area and the second attribute area each area of the image obtained by reducing the second attribute area is included. A means for encoding the data, a means for determining a single density information for each of the reduced second attribute areas, which is defined by the attribute data, and encoding the density information. An image information coding apparatus, comprising: a means for coding the first attribute area. 5. An image information decoding apparatus for decoding the image information encoded by the image information encoding apparatus according to claim 4, wherein: means for decoding the encoded attribute data; Means for modifying the attribute data so as to expand the second attribute area specified by the attribute data; means for decoding encoded density information for each of the second attribute areas; and the modified attribute data and Means for reproducing pixel information of the second attribute area based on the decoded density information; means for decoding the encoded first attribute area to reproduce pixel information of the first attribute area An image information decoding apparatus comprising: means for synthesizing the reproduced pixel information of the first attribute area and the reproduced pixel information of the second attribute area.