JP6024363B2 - Image processing device - Google Patents

Description

The present invention, DCT is relates to an image processing apparatus which performs image area separation using (Discrete Cosine Transform) transformation. The DCT transform is also called a discrete cosine transform, and is a method of orthogonal transform that compresses an image signal by removing spatial redundant components of the image signal.

  Documents handled by an image processing apparatus such as a digital copying machine include character originals, halftone originals, and photo originals. When these originals are reproduced by the image processing apparatus, the image quality required for the respective documents is different, so that each region is optimally processed by the image region separation processing. Various methods have been proposed for such image area separation processing. For example, Patent Document 1 provides an image area separation method that divides an image into blocks and performs image area separation of halftone dots and characters with high accuracy using DCT conversion.

  However, the conventional image area separation process can separate only two characters and halftone dots. On the other hand, recently, there are many images in which photographs and characters are mixed. Recently, a method of giving glossiness with a clear toner is often used for photographs.

  An object of the present invention is to divide an image into blocks and to perform image area separation of halftone dots, characters, and photographs with high accuracy using DCT transform.

In order to solve the above-mentioned problems, an image processing apparatus comprising an image input means for inputting an image, an image processing means for processing the input image, and an image output means for outputting the processed image is provided. Further, a dividing unit that divides the image data input from the image input unit into blocks, a DCT converting unit that performs DCT conversion on the pixels of each block, and a frequency component excluding the DC component among the DCT converted frequency components. The zigzag scan unit that performs zigzag scanning, the zigzag rate calculation unit that calculates the value obtained by doubling the data of interest and subtracting the data before and after it to check the peak of the scanned data, and the division unit Average luminance calculation unit for calculating the average of the luminance of the blocks obtained, and each block calculated by the zigzag rate calculation unit A low frequency band, high-frequency band and the absolute value of the sum and the character blocks each block by using the average luminance of each block calculated by the average luminance calculator of the zigzag rate in all frequency bands, screen dot block or photo A determination unit that determines one of the blocks, and a processing unit that performs smoothing or edge enhancement according to whether each of the determined blocks is a character, a halftone dot, or a photograph. Here, the low frequency band is, for example, the low frequency quarter portion of the entire frequency band excluding the DC component (about 2 to 16 on the horizontal axis in the graphs of FIGS. 4 to 6), and the high frequency band is, for example, the entire frequency band. It can be defined that the frequency is between 1/2 and 1/4 of the entire frequency band in the high frequency direction (about 33 to 48 on the horizontal axis in the graphs of FIGS. 4 to 6).

  According to the present invention, an image is divided into small blocks, and 64 coefficients output by DCT conversion of each block are zigzag scanned to obtain zigzag rates in low frequency components and high frequency components. Since the point image portion, the character image portion, and the photograph portion are separated, the image area can be separated with high accuracy.

It is a block block diagram concerning the Example of this invention. It is a figure explaining DCT conversion. It is a figure which carries out the zigzag scan of the coefficient block by which DCT conversion was carried out. It is a figure which shows the frequency characteristic of a character image. It is a figure which shows the frequency characteristic of a halftone image. It is a figure which shows the frequency characteristic of a photograph image. It is a flowchart of the image area separation process in the premise structure of this invention . It is a flowchart of the image area separation process in the Example of this invention . It is a figure which shows the example which corrects a separation result by seeing a surrounding block.

First, the premise structure of this invention is concretely demonstrated with drawing .
FIG. 1 is a block diagram common to the embodiments of the present invention. The upper part is a part for calculating an image area separation result, and the lower part receives an image area separation result, and an image corresponding to the result is obtained in block units. This is the part sent to the image output unit. In FIG. 1, an image input unit 1 has a photoelectric conversion element such as a CCD element, reads a document as image information, separates it into three colors of RGB, and outputs an 8-bit digital signal. The DCT conversion unit 8 performs DCT conversion on the image data (G component) of 8 × 8 pixels output from the image input unit 1. In this example , the G component is used. However, the RGB image is subjected to L * a * b conversion, and data representing the brightness of the image such as the L * component (brightness component: brightness considering human visual characteristics) is used. It may be used. The zigzag scanning unit 3 scans 63 frequency components excluding DC components (details will be described later). The zigzag rate calculation unit 4 calculates the peak degree of the 63 pieces of data (details will be described later). Based on the zigzag rate calculated by the zigzag rate calculation unit 4, the image area separation unit 5 separates the block of 8 × 8 pixels into character blocks, halftone blocks, and photo blocks, and the determination result is sent to the subsequent stage. hand over.

  The gamma correction units 6a to 6c correct the RGB images shifted by the image input unit 1 to the original RGB using a table prepared in advance for each pixel. When a document is read by a photoelectric conversion element such as a CCD element, the high-frequency component becomes small and the image becomes blurred. In order to correct it, the MTF correction units 7a to 7c restore the original high-frequency component. Apply high-pass filter to return to. The color correction unit 8 converts the input RGB into YMCK suitable for the output device. The smoothing processing unit 9 applies a low-pass filter to suppress moire. This is the part that creates data for halftone dots. Reference numeral 10a denotes an edge emphasizing unit 1, which applies a strong high-pass filter for emphasizing the edge of a character part. This improves the sharpness of the characters. This is the part that creates data for character blocks. Reference numeral 10b denotes an edge emphasizing unit 2, which applies a high-pass filter having a higher degree of raising a high frequency component than the edge emphasizing unit 1 (10a) in order to reproduce the sharpness of a photograph portion. Reference numerals 11a to 11d denote dither processing units that perform conversion processing so as to produce pseudo gradations. The selector 12 selects halftone dot block data, character block data, or photographic block data in accordance with the determination result from the image area separation unit 5, and outputs YMCK image data in units of blocks. If the image area separation result does not indicate any character, halftone dot, or photograph, the image is sent from the color correction unit to the dither processing unit without applying a filter (processing specific to the character, halftone dot, or photo block). Select and output the selected data. By outputting without performing the filtering process in this way, it is possible to prevent image deterioration due to erroneous determination. Reference numeral 13 denotes an image output unit such as a color printer.

The DCT converter 2 divides the pixels constituting the image into 8 × 8 pixel blocks, and converts the 8 × 8 pixel blocks from gradation (luminance) to frequency. The value of each pixel is expressed as f (i, j) (where i = 0 to 7, j = 0 to 7), and the mapping F (u, v) by DCT is
F (u, v) = (2 · C (u) · C (v) / N) ΣΣf (i, j) · cos {(2i + 1) uπ / 2N} · cos {(2j + 1) vπ / 2N}
It is expressed. The result of this two-dimensional DCT is called a DCT coefficient and is given by an 8 × 8 coefficient matrix (FIG. 2). However, the product-sum operation is performed from i, j = 0 to 7. Further, N = 8, u, v = 0, 1-7, C (w) = 1 / √2 (when w = 0) or 1 (when w = 1, 2-7).

  Since the result of DCT conversion in the zigzag scanning unit 3 is a two-dimensional matrix of 8 × 8 horizontal / vertical spatial frequency coefficients, it is converted into one dimension by a method called zigzag scanning. As shown in FIG. 3, the 63 frequency components excluding the DC component are scanned obliquely in the direction of the high frequency component starting from the low frequency portion. When it comes to the periphery of the two-dimensional matrix, it moves to one high frequency portion and scans again obliquely (numbers in the figure indicate the scan order).

  Then, the concept of zigzag rate is introduced and used for determination of character blocks, halftone blocks, and photo blocks. That is, the operation is performed by the zigzag rate calculation unit 4 in order to examine the peaks of 63 data obtained by the zigzag scan, and is performed according to the following procedure.

  The value (zigzag rate) obtained by doubling the data of the portion to be checked whether or not it is a peak and subtracting the data before and after that is obtained. 61 pieces of data excluding the points (u, v) = (0, 0) and (7, 7) are obtained. That is, zigzag rate [i] = zigzag [i] × 2-zigzag [i−1] −zigzag [i + 1] (1 ≦ i ≦ 63)

  The image area separation unit 5 is a part that uses the data from the zigzag rate calculation unit 4 to determine one of a character block, a halftone dot block, and a photo block, and sends the determination result to the selector 12 and the image output unit 13. . If it is determined that the block is a photographic block, the image output unit 13 adds CMYK toner to give gloss. If no character block, halftone block, or photo block is determined, the determination result is sent to the image output unit 13 and the selector 12 as an unknown block.

  The image area separation of characters, halftone dots, and photographic blocks using zigzag rate is performed as follows. That is, when DCT conversion is performed in block units, the character image, the halftone dot image, and the photographic image have different frequency characteristics as shown in FIGS. 4 shows the frequency characteristics of the character image, FIG. 5 shows the frequency characteristics of the halftone dot image, and FIG. 6 shows the frequency characteristics of the photographic image (the horizontal axis is the scan position, and the vertical axis is the DCT coefficient). That is, many peaks are likely to appear in the low frequency component in the character image, many peaks are likely to appear in the high frequency component in the halftone image, and no peak appears in the photographic image. Therefore, in the sum of absolute values of zigzag rate, the character image, the halftone dot image, and the photographic image are different, and the character image shows a high value in the low frequency portion and a low value in the high frequency portion. On the other hand, the halftone image is the opposite, and its value is small in the low-frequency part and large in the high-frequency part. In a photographic image, the sum of zigzag rates in all frequency bands is smaller than that of a character image or a photographic image.

  Therefore, the 63 frequency components excluding the DC component are scanned obliquely in the direction of the high frequency component starting from the low frequency portion. In order to examine the peaks of the 63 data obtained in this way, the data of the portion to be examined whether or not it is a peak is doubled, and a value (zigzag rate) obtained by subtracting the data before and after that is obtained. Conditions for capturing the features related to the sum of absolute values of zigzag rates in the character block, halftone dot block, and photo block are set, and each block is determined. Such image area separation processing is performed by performing DCT conversion on the divided blocks.

FIG. 7 is a flowchart showing an example of the image area separation processing of the present configuration . First, an 8 × 8 pixel block is captured (S1). Next, the fetched block is DCT-converted into frequency components of 8 × 8 = 64 coefficients (S2). Next, the 63 coefficient frequency components excluding the DC component are zigzag scanned to obtain 61 zigzag rates from the zigzag scan data (S3). Next, the sum SumZrateLow of the absolute value of the zigzag rate in the low frequency band is calculated (S4), the sum SumZrateHigh of the absolute value of the zigzag rate in the high frequency band is calculated (S5), and the absolute value of the zigzag rate in the entire frequency band is calculated. Sum Sumrate All is calculated (S6).

  Next, it is determined whether or not the sum of the absolute values of the zigzag rate in the low frequency portion is smaller than a predetermined value th1, and whether or not the sum of the absolute values of the zigzag rate in the high frequency portion is larger than a predetermined value th2 ( S7) When the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th1, and the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th2, the block is determined to be a halftone block, An instruction to select the block to which the smoothing filter has been applied by the selector is issued (S8).

  If it is not determined as a halftone dot block, it is determined whether or not the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th3, and the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th4. (S9), when the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th3 and the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th4, the block It is determined that the block is a character block, and an instruction to select a block with strong edge emphasis is issued by the selector (S10).

  If neither the character block nor the halftone dot block is determined, the block is determined to be a photographic block, and an instruction to select a weakly edge-enhanced block with a selector is issued (S11). In addition, the image output unit issues an instruction to apply clear toner (S12). Glossy photographic images can be output by applying clear toner.

In this example , the determination is made in the order of halftone dot block → character block → photograph block.

Example 1
This is an example in which an aspect when the determination result is unknown is also added. The description regarding FIGS. 1 to 6 is the same as that of the premise configuration . By the way, one of the causes of the separation error is to divide it into three parts: a character part, a halftone part and a photograph part. In other words, it is because the separation process is performed even on a portion that is difficult to determine. The fatal error is reduced by outputting the input image as it is without performing the filtering process, rather than separating the difficult part. Therefore, in this embodiment, an instruction not to perform edge enhancement or smoothing filter processing is given to blocks that are difficult to separate.

  FIG. 8 is a flowchart illustrating an example of image area separation processing according to the present exemplary embodiment. First, an 8 × 8 pixel block is captured (S1 ′). An average luminance value AveKido of the block is calculated (S2 '). Next, the fetched block is DCT-converted to a frequency component of 8 × 8 = 64 coefficients (S3 ′). Next, the 63 coefficient frequency components excluding the DC component are zigzag scanned to obtain 61 zigzag rates from the zigzag scan data (S4 '). Next, the sum SumZrateLow of the absolute value of the zigzag rate in the low frequency band is calculated (S5 ′), the sum SumZrateHigh of the absolute value of the zigzag rate in the high frequency band is calculated (S6 ′), and the absolute zigzag rate of the entire frequency band is calculated. The sum SumZrateAll of values is calculated (S7 ′).

  Next, it is determined whether or not the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th1, and it is determined whether or not the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th2 (S8). ') When the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th1 and the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th2, the block is determined as a halftone block, An instruction to select the block to which the smoothing filter has been applied by the selector is issued (S9 ′).

  If it is not determined as a halftone dot block, it is determined whether or not the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th3, and the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th4. If the sum of the absolute values of the zigzag rate in the high frequency part is larger than the predetermined value th3 and the sum of the absolute values of the zigzag rate in the low frequency part is smaller than the predetermined value th4, the block Determines that the block is a character block, and issues an instruction to select a strongly edge-enhanced block with a selector (S11 ′).

  If neither the character block nor the halftone dot block is determined, the average luminance AveKido of 8 × 8 pixels is smaller than the predetermined value th5, and the sum of the absolute values of the zigzag rate in all frequency bands is smaller than the predetermined value th6. At this time, the block is determined to be a photograph block (S12 ′), and an instruction to select a weakly edge-enhanced block with a selector is issued (S13 ′). In addition, the image output unit issues an instruction to apply clear toner (S14 ').

  A block that has not been determined to be a character block, halftone block, or photographic block issues an instruction to select a block that is not subjected to filter processing with a selector as a block that cannot be determined (S15 ').

In this embodiment, the determination is made in the order of halftone dot block → character block → photograph block. However, the present invention is not limited to this, and halftone dot block / character block / photograph block may be determined in any order.
In this embodiment, the first to sixth predetermined values are determined, the character block, the halftone block, and the photo block are determined under the respective conditions, and selection signals for the respective determination images are output to the selector 12. .

(Example 2 )
This is an example of correcting the determination result with reference to surrounding block information. Since image area separation is performed with reference to surrounding block information, it is possible to prevent a separation error that abruptly makes an erroneous determination from a certain block, thereby realizing a highly accurate determination. 1 to 6 and 8 are the same as those in the first embodiment . In this embodiment, information on the surrounding four blocks (upper, left, two left, and diagonally upper blocks of the target block) is incorporated as a key in the constant part of the above-described determination condition. Since the blocks are determined in raster order, the right block and the lower block of the block of interest have not been determined yet and cannot be used.
key = 0.25 (flag (upper) + flag (left)) + 0.125 (flag (two left) + flag (diagonally upper right) + flag (diagonally upper left))
Here, flag takes a value of -1 for characters, 1 for halftone dots, and 0 for photographs (FIG. 9).

  By incorporating the key into the condition, a block with a surrounding character is easily determined as a character, a block with a halftone dot is easily determined as a halftone dot, and a block with a surrounding photo is easily determined as a photograph.

  When the key is incorporated in the determination using the zigzag rate described above, the result is as follows. That is, when the sum of the absolute values of the zigzag rate in the low frequency portion <th1 + key × constant a and the sum of the absolute values of the zigzag rate in the high frequency portion> th2 + key × constant b, it is determined as a halftone image. The constant a is a positive constant and has an absolute value smaller than th1. The constant b is a negative constant and has an absolute value smaller than th2. When the sum of absolute values of zigzag rate in the high frequency part <th3 + key × constant c and the sum of absolute values of zigzag rate in the low frequency part> th4 + key × constant d, it is determined as a character image. The constant c is a negative constant whose absolute value is smaller than th3. The logarithm d is a positive integer and has an absolute value smaller than th4.

  When the average luminance AveKido <th5 of 8 × 8 pixels and the sum of absolute values of zigzag rate in all frequency bands ≦ th6 + key × constant e, it is determined as a photographic image. The constant e is a positive constant and the absolute value is smaller than th6.

  The constants a, b, c, d, and e are set so that the separation accuracy is improved under the above conditions. About the magnitude | size of an absolute value, the constants a, b, c, and d set the substantially same value, and the absolute value of the constant e is set to a value smaller than the constants a, b, c, and d.

  It is also conceivable that the above image processing method is independently recorded on a recording medium and read by a CPU in an image forming apparatus image such as a copying machine, a facsimile, or a printer.

DESCRIPTION OF SYMBOLS 1 Image input part 2 DCT conversion part 3 Zigzag scanning part 4 Zigzag rate calculation part 5 Image area separation part 6 Gamma correction part 7 MTF correction part 8 Color correction part 9 Smoothing processing part 10 Edge emphasis part 11 Dither processing part 12 Selector 13 Image output unit

JP-A-6-303425

Claims (5)

An image processing apparatus having an image input means for inputting an image, an image processing means for processing the input image, and an image output means for outputting the processed image is input to the image processing means from the image input means. A division unit that divides the image data into blocks, a DCT conversion unit that performs DCT conversion on the pixels of each block, a zigzag scanning unit that performs zigzag scanning of frequency components excluding the DC component among the DCT converted frequency components, In order to examine the peak of the scanned data, the attention data is doubled, and a zigzag rate calculation unit that calculates a value obtained by subtracting data before and after that as a zigzag rate, and an average of the luminance of the blocks divided by the division unit An average luminance calculation unit to calculate, and a low frequency band, a high frequency band and a block of each block calculated by the zigzag rate calculation unit The sum of the absolute values of the zigzag rate in all frequency bands, determining determines that one of the average character block of each block by using the average luminance of each block calculated by the luminance calculation unit, screen dot block or photo block And a processing unit that smoothes or edge-enhances each determined block depending on whether it is a character, a halftone dot, or a photograph.   When the sum of absolute values of zigzag rate in the low frequency band is smaller than the first predetermined value and the sum of absolute values of zigzag rate in the high frequency band is larger than the second predetermined value, the target block is determined as a halftone block, and the high frequency band The block of interest is determined to be a character block when the sum of the absolute values of zigzag rate in is lower than a third predetermined value and the sum of the absolute values of zigzag rate in the low frequency band is greater than a fourth predetermined value. Item 8. The image processing apparatus according to Item 1.   When the average luminance of the block is smaller than the fifth predetermined value and the sum of the absolute values of the zigzag rate in all frequency bands is smaller than the sixth predetermined value, the target block is determined as a photo block, and a halftone dot block, a character block, and a photo block The image processing apparatus according to claim 2, wherein neither of smoothing nor edge enhancement is performed because the determination is impossible when neither of the determinations can be made.   The image processing apparatus according to claim 3, wherein when it is determined that the block is a photograph block, clear toner is applied to the block.   The image processing apparatus according to claim 1, wherein the determination result of the target block is corrected with reference to determination information of blocks around the target block.

Images