JPH04144485A - Image processor - Google Patents

Abstract

PURPOSE:To attain highly developed compression while holding high picture quality by separating an objective document image into a half tone area and a character area and using a coding system appropriate for the statistic properties of respective areas. CONSTITUTION:An input image is separated into a character area and a half tone area by an area sorting means 11 and the character area is divided into plural blocks by a blocking means 12. The 1st coding means 13 codes respective blocks by a coding system appropriate for the statistic property of the character area and the 2nd coding means 14 executes coding by a coding system appropriate for the statistic property of the half-tone area. Since the image is divided into two areas and coded by means of the coding systems appropriate for respective areas, highly developed compression can be attained while holding high picture quality and the processing time can also be shortened.

Description

[Detailed description of the invention]

[Industrial Field of Application] The present invention relates to an image processing apparatus that encodes a multivalued image including a character region and a halftone region, or decodes a code encoded in this manner.

[Conventional technology]

Images are information that can intuitively control the whole picture, and community communication using images has become widely popular as an extremely excellent means of transmitting information. However, since image information generally has an extremely large amount of information,
If this original image information is transmitted to a remote location through a communication network or stored in various electronic media in the form of original image information, cost and processing time become major constraints. For this reason, a high-efficiency encoding method for efficiently compressing, transmitting or storing image information is required. High-efficiency coding methods include predictive coding and block truncate coding (BTC).
ionCoding), orthogonal transform coding method, vector quantization method, entropy? The No. 4 method is known.

[Problem to be solved by the invention]

Various high-efficiency encoding methods have been proposed, but each has advantages and disadvantages depending on the nature of the target document image. It wasn't enough. SUMMARY OF THE INVENTION An object of the present invention is to provide an image processing device that can encode/decode image information in a short processing time while maintaining high image quality and highly compressing the image information. be.

[Means to solve the problem]

The present invention is an image processing device for encoding a multivalued image, and includes a classification means (see FIG. )
As a result of the classification by this classification means, λl is classified into the character area.
Blocking means (first
12), a first 9n encoding means (FIGS. 1 and 3) that encodes each block formed into blocks by this blocking means, and a halftone area as a result of the classification by the classification means. A second encoding means (FIG. 14, FIG. 5) that performs 781 encoding on the classified area is implemented. Further, the image processing device for encoding a multivalued image of the present invention includes:
In order to encode the halftone area, there are a sampling means (FIG. 5, 51) that takes representative values of the input image at regular intervals, and an approximation means (51) that approximates the input image to a curved surface from the sampling result.
Fig. 5 53) and calculation means for calculating an approximation error by comparing the input image and the approximation result of the approximation means (Fig. 5 54)
If the reduction result of this calculation means is smaller than a predetermined threshold, it is encoded as is, and if it is larger, additional information is added (=1
Add and encode? n encoding means (FIG. 5, 55 to 58). Further, the device for decoding the result encoded by the image processing device of the present invention includes means (111 in FIG. 11) for classifying the encoded data into “ima” button information and curved surface approximation information;
If there is the additional information, the means for correcting the restored image using the additional information (114 to 117 in FIG. 11)
). Further, according to one aspect of the present invention, the approximation means (FIG. 53
) changes the size of the curved surface according to the density change of the image,
If the changes are dense, the curved surface is made smaller, and if the changes are monotonous, the curved surface is made larger (-4 composition can be used).

[Operation] The input image is separated into a character area and a halftone area by the area classification means, and the character area is divided into blocks by the blocking means. The first 9X1 encoding means encodes each block using an encoding method suitable for the character area standard and II objective quality. Further, the second encoding means performs encoding using an encoding method suitable for the uniform characteristics of the halftone region. In this way, the present invention divides an image into regions and initial encodes them using an encoding method suitable for each region, so it is possible to perform a high degree of compression with high image quality and reduce processing time. can. Furthermore, in the 1n encoding of the present invention, sampling is performed by the sampling means and the pixel density of the sample point is encoded, so that the amount of code can be highly compressed. Furthermore, when the difference (approximation error) between the image interpolated from the sample points by the approximation means by curved surface approximation and the human image is larger than a predetermined threshold, the encoding means encodes the approximation error as additional information, and the encoding means encodes the approximation error as additional information. Add to the density encoded information. Therefore, since errors can be corrected using the A button information during decoding, encoding and decoding can be performed with high quality. In addition, the present invention provides efficient sampling by making the punch size for curved surface approximation variable by performing fine sampling for areas with rapid density changes and coarse sampling for monotonous areas, depending on the nature of the image. Encoding becomes possible. [Embodiment] Fig. 2 shows a schematic 111X configuration of an image processing apparatus in which the present invention is applied to a digital copying machine, in which an image input unit 21 such as an image scanner, image conversion, editing, and other image processing are performed. an image processing unit 22 that performs image processing, an image output unit 23 such as a printer, and a code encoding/decoding unit (C
It consists of an ODEC) 24 and a storage section 25 that stores encoded images. FIG. 1 shows the configuration of an embodiment for encoding according to the present invention, including an area classification unit 11 that classifies an input image into two areas: a character area consisting of character figures, etc., and a halftone area, such as a photograph. , a blocking unit 12 that divides the input image into blocks of N1×N2 (N1 and N2 are positive integers) pixels for areas classified as character areas as a result of the classification by the area classification unit 11; 12, and a region classification unit 11.
As a result of the classification, the halftone region encoding unit 14 encodes the regions classified as halftone regions. The area classification unit 11 and the blocking unit 12 are the image processing unit 22 in FIG.
The character area encoding section 13 and the halftone area encoding section 14 belong to the function of the encoding/decoding section 24 shown in FIG. The document image input by the image input unit 21 is divided into a text area consisting of sentences, graphs, line diagrams, etc., and a halftone area including diagrams and photographs expressing shading using halftone dots and dithering by the area classification unit 11. To separate. The separation may be performed automatically using the properties of the image, or may be instructed by an operator. The blocking unit 301 divides the character area into blocks of N1×N2 pixels. In this embodiment, a 4×4 pixel block is used. Each block is highly efficiently encoded by the character area encoding section 13 and stored in the storage section 25. Here, the character area encoding unit 13 will be explained in detail. FIG. 3 is a diagram showing details of the character area encoding section 13 in FIG. 1. The character area encoding unit 13 includes a block memory 302 that stores blocks of character areas cut out by the blocking unit 301 (FIG. 2, 12), and an average density value calculation unit 303 that calculates the average density value within the block. , a standard deviation calculation unit 304 that calculates the standard deviation within a block, a label determination unit 305 that determines the nature of the image based on the calculated average density value and standard deviation value, and determines the nature of the image based on the average density value and standard deviation value. As a result of the determination by the determination unit 305, if the block belongs to a part where the density changes are very gradual (the [background part] of the label), the one-tone block approximation encoding unit (one-tone block approximation encoding unit) used to encode the block BTC) 306, and a two-tone block approximation encoding unit (two-tone BTC) 307 used to encode the block when the block belongs to a relatively gradual density change (label ■ [semi-background area]). , and when the block belongs to a part with a large density change (label ■ [character part]), the linear quantization unit 308 performs linear quantization, and the 1st gradation BTC 306, the 2nd gradation BTC 307, or the linear quantization unit 308 performs linear quantization. A Huffman encoding unit 309 that compresses and encodes gradation information, 1 gradation BTC 306, 2 gradation BTC 30
7 and linear quantization section 308, and a code assembling section 311 that constructs the initial code as illumination. Next, the operation of the character area encoding section will be explained. The block of the character area stored in the block memory 302 has an average density value Av
(11 deviation calculation unit 304 calculates the standard arc 1 difference S
Calculate d. The label determination unit 305 determines the average density value A V
e % standard deviation Sd1 and appropriately determined threshold Th
l, Th2 to 11. Then, the label is determined as follows. Label ■ (background part): Ave>Thl Label ■ (semi-background part): Ave≧Th1 and Sd<Th2 Label ■ (text part): Ave≧Th I s and Sd
≧Tri2 When the label determining unit 305 determines that the label is ■, the block is a flat portion with gradual density changes. Based on this uniformity, the inside of the block is approximated by one gradation based on one quality. Therefore, 1 gradation block approximation? ] Using the encoding unit 306,
The average density value within the block is taken as the gradation of the block. If the label determination unit 305 determines that the block is labeled ■, the density change in that block is not flat, but is gradual. Based on this systematic property, the inside of the block is approximated by two gradations. Therefore, the 2-gradation block approximation right-signing unit 307 is used to generate gradation information X consisting of two density representative values. , xl and
Resolution information S]l indicating which representative value each pixel is approximated by is determined. When the label determining unit 305 determines that the block is labeled as ■, the block is a portion where the density changes rapidly. Linear quantization is used as a process suitable for this statistical property. That is, the linear quantization unit 308 sets a level for the background portion, and performs linear quantization using a 166-gradation quantization level, which is obtained by dividing the other portion into 15 equal parts and setting the level. The Huffman initial coding unit 309 uses a Huffman code to compress the gradation information output from the 1st gradation block approximate encoding unit 306, the 2nd gradation block approximate encoding unit 307, and the linear quantization unit 308 by assigning a variable length code. make changes. Furthermore, the MMR encoding unit 310 assigns O to all pixels for the output from the 1-gradation block approximation encoding unit 306 with the label ■, and outputs the output from the 2-gradation block approximation encoding unit 307 with the label ■. Based on the obtained resolution information, 1 is assigned to all pixels for the output from the linear quantization unit 308 with the label ■, and the output is collectively MMRn encoded. ? 'lhi! II-γ section 311 is Huffmano initialization section 309
And an encoded output is generated based on the output of the MMR encoder 310. FIG. 4 shows a code structure assembled to store a character area of -. in the storage section 25, and encodes information of one character area. To/soda information H; both image rise of the sentence ma' area and area position label 1 ■: gradation information HJ G hill bell ■: gradation information G and resolution information P-/beno 1 (■ "g)) sub information The character image encoded as described above is stored in the storage unit 25.Next, the information initialized and stored for the character area is restored from the gradation information and resolution information. This is done by reproducing the density distribution of the block.For blocks labeled ■ and label ■, each block has 1 gradation and 2 gradations (representative values)
Play with. Regarding the label ■, the density value of each pixel of the block is reproduced using 16 representative values. Next, processing of the intermediate j11 area will be explained. FIG. 5 shows an example of the configuration of the halftone area encoding section 14 in FIG.
2. Curved surface approximation unit 53, approximation error R1 calculation unit 54, nonlinear quantization unit 55, Huffman encoding unit 56, mode determination unit 5
7 and a sign assembly section 58. In the encoding algorithm in this example, first, the sampling unit 51 performs sampling in a quincunx pattern to utilize the correlation between density values in the vertical, horizontal, and diagonal directions. This→
Since the non-bringing position [γt is fixed, information representing the position of the sample point is unnecessary. Next, the density values of the sample points are predictively encoded by the predictive coding unit 52 in order to reduce the amount of code since the variance of the density values at adjacent sample points is small. Then, the approximation target area surrounded by four adjacent sample points is approximated by the curved surface approximation unit 53 using a bicubic patch, and the approximation error is calculated as the mean square difference between the original image and the approximation image within the area. The area is determined by the section 54 for each area. If the mean square error is a threshold value, the mode determination unit 57 compresses and encodes the approximation error (additional information) of each pixel within the region by the nonlinear quantization unit 55 and the Huffman encoding unit 56. Each process will be explained in more detail below. The sampling section 51 performs sampling in a quincunx pattern as shown in FIG. 6 in order to utilize the correlation of density values in the vertical, horizontal, and diagonal directions. The interval between the sample points is set to 4 pixels in order to minimize blurring in areas where density changes are significant during approximation. Furthermore, the density value of the sample point is the average density value of the four pixels indicated by diagonal lines in FIG. Sample points are located not at pixels but at boundaries between pixels. The predictive encoding unit 52 uses the average value of four reference sample points adjacent to the sample point of interest shown in FIG. 7 as a predicted value, and subtracts this predicted value from the density value of the sample point of interest to obtain a prediction error. . This prediction error is subjected to nonlinear quantization and approximated using 19 representative values. Next, each representative value is assigned a Huffman code and encoded. The curved surface approximation unit 53 forms an area (
A process of approximating a diamond-shaped region having four vertices (hereinafter referred to as an approximation target region) by applying a bicubic batch is performed. In the bi3vζ patch, the approximation target region is interpolated with reference to the 16 sample points shown in FIG. The approximation error d1 calculation unit 54 calculates the mean squared error MSE between the original image and the approximation image for each area, as shown in FIG. to ask. The mode determination unit 57 determines that if the mean squared error MSE is less than or equal to the threshold TH, it is mode + ((=region where 1 additional information is required), and if it is less than the threshold 78, it is determined to be mode 2 (region where no additional information is required). Here, the mote information for creating the mode information MD indicating which mode the approximation target region belongs to is a binary image of 0.1, so when encoding, MM
R encoding is adopted. When the mode determination unit 57 determines mode 1, the approximation error is encoded by the nonlinear M-coding unit 55 and the Huffman encoding unit 56. In this embodiment, the nonlinear quantization unit 55 approximates using 11 representative values. Then, the Huffman encoding unit 56 assigns a Huffman code to each representative value and initializes it. The code assembling unit 58 assembles the prediction error S of the sample point, the mode information M, and the approximation error of each pixel in the approximation target area, that is, the additional information F, into a code configuration as shown in FIG. 10, and generates an encoded output. do. Header information H is inserted at the beginning of the initial encoded output. The header information H is information such as the order 1ξ and size of the area when encoding an image in the halftone area. Note that the additional information is necessary only in mode 1. As described above, decoding of encoded halftone region images is as follows:
For example, it can be executed by a decoding section having an I4 configuration as shown in FIG. The information separation unit III separates prediction error information S1, additional information F1, mode information M, etc. from the code (M1-1 information) as shown in FIG. 117. Based on the density information of the sample points output from the decoding unit 112 of the predictive coding code, the curved surface approximation unit 113 interpolates by bicubic batch to obtain the density value of each pixel. In this case, the selection unit 118 selects and outputs the output of the curved surface approximation unit 113 as it is as a decoded output.When the mode information is mode l, the correction unit], l5 selects and outputs the output of the curved surface approximation unit 113 in the additional information decoding unit 114. The correction is performed using the decoded approximation error value, and the selection unit 116 selects and outputs the output of the correction unit 116 as the decoded output.In the following embodiment 1- (first embodiment), the patch size of the halftone area is Although we have explained the case where the sample points are set at fixed intervals, it is also possible to make the sample points variable according to the properties of the image and change the patch size. As an example of 2, we will explain a method of varying the patch size (therefore, sample points). In this method, the image to be encoded is divided into a number of rhombuses (slanted squares), or patches, and each patch The points necessary to construct the sample point are extracted as sample points, and the density information of these sample points is stored.In decoding, the density value within the patch is stored based on the density information of this sample point. The image is restored by doing this for all patches.The method for determining the patch size is that halftone images contain parts with sharp density changes, such as the surface of an object, and areas with a monotonous density distribution, such as the background. Efficient encoding is possible by performing fine sampling in regions with rapid changes in the third degree and coarse sampling in monotonous regions.A second embodiment is shown in Fig. 5. First embodiment and configuration (1m
1. The difference is mainly in the sampling section 51.
The method of selecting sample points in is based on the patch size, and the other configurations are almost the same. As shown in FIG. 12, the patch is a diamond-shaped area with four sample points as vertices, and is a square that is tilted at 45 degrees with respect to the pixel arrangement. Further, as in the first embodiment, the sample point is virtually taken at the center of four pixels, and the average value of these four pixels is taken as the density value of the sample point. When the punch size is minimum, m is 2, and as shown in Figure 12, the patch is expanded into a bell shape with the topmost point of the patch fixed, and n is Determine from 0 to 7. The procedure for making that determination is as follows. (2) As shown in FIG. 13, density gradient information in the state of minimum patch size addition (n=o) is obtained as reference information. II) Divide n=1 punches into minimum patch sizes and find the density gradient for each minimum patch in the same way as the reference information. (2) If the absolute value of the difference between the absolute value of the density gradient and the absolute value of the reference information is greater than or equal to the threshold value (THI), then n=o is set as the patch size. If it is less than the threshold value, n=2 and the same process as - is performed. However, if the minimum patch that has already been patched is included, n=0 regardless of the density gradient. TV) n′? ! : Vary from 0 to 7 to determine the patch size. Once the patch size is determined, sample points can be extracted. Predictive coding is performed for each sample point. Predictive encoding can be performed in the same manner as described in the first embodiment. However, unlike the first embodiment, two problems arise because the patch sampling is performed while changing the patch size. In other words, 1) there are not always 4 tons used for prediction, 2) there are cases where there are no points at all. Taking this into consideration, each ttil W point is processed as follows. Problem (3) is to calculate the average value using only existing points. Regarding the problem (2), the value of the vertex located in the second most part of the patch is used as the predicted value. In the case of the second row at the top of the screen, the vertex located at the top also has no value, so the median value of the usable density value range is used as the predicted value. The code structure has the structure shown in FIG. 10 plus information on the coordinates of the two-most vertices of the patch and the patch size.

【Effect of the invention】

According to the present invention, a target document image is divided into two regions, a halftone region and a character region, and an encoding method suitable for the statistical properties of each region is used, thereby achieving high image quality. It is possible to achieve a high degree of compression while maintaining Further, according to the present invention, when encoding a halftone region, a region surrounded by four sampled points is approximated by applying a bicubic patch, and when the approximation error is greater than a certain threshold, Since image error information is added, image quality can be improved. Furthermore, in an embodiment of the present invention in which the sampling points are finely sampled in areas where the image density changes sharply and coarsely sampled in monotonous areas in encoding the halftone area, - layer efficient encoding is possible.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an encoding section of an image processing apparatus according to an embodiment of the present invention. FIG. 2 is a diagram showing a general configuration of an image processing apparatus in which the present invention is applied to a digital copying machine. FIG. 3 is a diagram showing an example of the configuration of the character area encoding section in FIG. 1. FIG. 4 is a diagram showing the code structure of the code output from the character area encoding section. FIG. 5 is a diagram showing an example of the configuration of the halftone area encoding section in FIG. 1. FIG. 6 is a diagram showing sample points for fixed density sampling. FIG. 7 is a diagram showing reference sample points for predictive encoding. FIG. 8 is a diagram showing the coordinate form of a bicubic patch. FIG. 9 is a diagram for explaining the mean square error between the original image and the approximate image within a region. FIG. 10 is a diagram showing a code structure for encoding a halftone region. FIG. 11 is a diagram showing a configuration for decoding a halftone region. FIG. 12 is a diagram for explaining how to determine the patch size. FIG. 13 is a diagram for explaining the concentration gradient.

Claims (4)

[Claims] (1) An image processing device that encodes a multivalued image, which includes a classification means for classifying an input image into two regions: a character region consisting of characters, figures, etc., and a halftone region such as a photograph, and a classification means for classifying the input image into A blocking means for dividing an input image into blocks of N1×N2 (N1 and N2 are positive integers) pixels for regions classified as character regions as a result of the classification, and each block formed by the blocking means. An image processing apparatus comprising: a first encoding means for encoding; and a second encoding means for encoding a region classified as a halftone region as a result of classification by the classification means. (2) An image processing device for encoding a multivalued image, comprising: sampling means for taking representative values of an input image at regular intervals; approximation means for approximating the input image to a curved surface from the sampling results; and the input image and the approximation. A calculation means that calculates an approximation error by comparing the results of the approximation of the calculation means, and an encoding method that encodes the calculation result of the calculation means as it is if it is smaller than a predetermined threshold, and adds additional information and encodes it if it is larger. An image processing device comprising: means. (3) A device for decoding the result encoded by the image processing device according to claim (2), and means for classifying the encoded data into additional information and curved surface approximation information;
An image processing apparatus comprising: means for correcting a restored image using the additional information, if the additional information is present. (4) The approximation means changes the size of the curved surface according to the density change of the image, and when the change is drastic, the curved surface is made smaller, and when the change is monotonous, the curved surface is made larger. 2) The image processing device described above.