JPH1153537A - Method for restored image interpolative and corrective filtering - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reproduce an original resolution image approximately with a smooth light and shade pattern by interpolating and generating spectral characteristics based on a multi-valued filter of low resolution and making corrections for reproducing the light and shade pattern. SOLUTION: An M×M extraction part 42 generates block data 43 of M×M pixels by dividing a dither image 41 and a 2D-DCT process part 44 finds the spectral value of a DCT area from the luminance value of M×M size to obtain spectral M×M block data 45. A filtering processing 47 for conversion to a multi-valued image is performed to obtain multi-valued spectral M×M block data 48. A decimation processing part 49 discards high-frequency components from the data 48 to obtain decimation spectral N×N block data 410. Further, a spectrum interpolation part 413 generates and interpolates spectral M×M data 414 from an interpolation coefficient from an interpolation coefficient analysis part 411 and the N×N block data 410 so as to restore the discarded high-frequency components.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a processing method for generating and restoring a multi-bit image or a high bit image such as a continuous tone image from a low bit image such as a binary image, and a reproducing method.

[0002]

2. Description of the Related Art With respect to image restoration in spatial frequency spectrum processing in general, it is assumed that a noise model spectrum is mixed in an original image spectrum with respect to a spectrum of an image quality degraded image. There are many methods for removing the deteriorated part formed by the noise model by multiplying the image quality deteriorated image by the inverse function.

FIG. 7 shows a configuration example of a conventional image restoration. In the figure, reference numeral 11 denotes an original image spectrum, 12 denotes an image quality deteriorated image spectrum, 13 denotes a restored image spectrum, 14 denotes a noise analysis processing unit, 15 denotes a noise pattern model, 16 denotes noise inverse characteristic data, 17 denotes noise addition processing, and 18 Represents noise removal processing.

[0004] In the case shown in FIG. 7, a noise pattern model 15 is added to an original image spectrum 11 in a noise adding process 17 to obtain an image quality degraded image spectrum 12. On the other hand, the noise pattern model 15 is analyzed by the noise analysis processing unit 14, and the noise analysis processing unit 14 obtains noise inverse characteristic data 16. The noise inverse characteristic data 16 is calculated with the image quality deteriorated image spectrum 12 in the noise removal processing 18. As a result, a restored image spectrum 13 without noise is obtained.

On the other hand, as one of methods for generating and restoring a high-quality multi-valued image from an image having a deteriorated image quality and a low-bit image, an image is subjected to spectral conversion and multi-valued in a frequency domain. There is a method of generating a desired multi-valued image by performing a filter response function and image filtering. As the filter response function characteristic, a low-pass filter type amplitude characteristic is generally used, and has an effect of suppressing a high-frequency component peculiar to a binary image, and thus is suitable for generating a multi-valued image.

FIG. 8 shows an example of multi-level image filtering. In the figure, reference numeral 21 denotes a binary image (low bit image such as a dither image), 22 denotes pixel M × M block data,
3 is a spectrum conversion process, 24 is a spectrum M × M block data, 25 is a multi-valued filter M × M block data, 26 is a multi-valued image filtering process, 27 is a decimation process, and 28 is a decimation spectrum N ×
N block data, 29 is a spectrum inverse conversion process, 21
0 represents pixel N × N block data, and 211 represents a multi-valued image.

The spectrum conversion processing 23 is performed on the block data 22 in the binary image 21, and the spectrum M
× M block data 24 is obtained. Multi-value conversion filter M × M block data 2 for the block data 24
5 is calculated in a multi-valued image filtering process 26, a decimation process 27 is performed, and a decimation spectrum N × N block data 28 is obtained. The block data 28 is subjected to a spectrum inverse conversion process 29
Is performed, and a multi-valued image 211 having pixel N × N block data 210 is obtained.

That is, after suppressing high-frequency components in a binary image, the suppressed high-frequency components are additionally subjected to decimation processing to generate a multi-resolution multi-valued image with an image size smaller than the size of the original image. It has been like that.

FIG. 9 is a diagram for explaining the decimation process. Multi-valued image filtering processing 26 in FIG.
Is performed, and the spatial spatial thinning is performed in each of the spatial frequency u direction and the spatial frequency v direction, and the decimation spectrum N × N block data 2
8 have been obtained.

[0010]

In the conventional case, a degraded image is dealt with by assuming an original image and a noise model. However, it is desired to restore a low-bit image having deteriorated image quality to an image having the original resolution without assuming the deteriorated image quality in the form of an original image and a noise model.

The present invention restores and reproduces an original resolution image from a low-bit image using a low-resolution multi-valued spectrum obtained from the low-bit image or the degraded image. The purpose is to approximately restore the smooth shading pattern that was possessed by.

[0012]

SUMMARY OF THE INVENTION According to the present invention, various images such as character images and natural images are binarized or half-toned to a low resolution (lower resolution than the original image). ), The spectral characteristics for obtaining the restored image are interpolated and generated based on the multi-valued filter, and the correction is performed to approximately reproduce the shading pattern unique to the original image, thereby restoring the deteriorated image. It is characterized by the following.

By using the method of the present invention, the binarized or halftoned image of the original image can be
Even without assuming a noise model, it is possible to generate a high-quality restored image using only the spectral values obtained from the image-degraded image.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following, a dither image obtained by dithering an original image into a low bit image is assumed, and a two-dimensional cosine transform (hereinafter, referred to as a spectral transform) is used for the spectral transform.
2D-DCT, and the inverse spectrum transform is referred to as 2D-IDCT).

[Embodiment 1] FIG. 1 shows a block diagram of image filtering according to an embodiment of the present invention.

In the drawing, reference numeral 41 denotes a dither image, 42 denotes a pixel M × M block data extracting unit, and 43 denotes a pixel M ×
M block data, 44 is M × M block 2D-DCT
A processing unit, 45 is dither spectrum M × M block data, 46 is a multivalued filter M × M block data, 47
Is a multi-valued image filtering process, 48 is a multi-valued spectrum M × M block data, 49 is a decimation processing unit, 410 is a decimation spectrum N × N block data, 411 is an interpolation coefficient analysis unit, and 412 is a spectrum N × N block. An interpolation coefficient for interpolating spectrum M × M block data from data, 413 is a spectrum interpolation processing unit, 414 is a restored spectrum M × M block data, 415 is an M × M block 2D-IDCT processing unit,
Reference numeral 16 denotes pixel M × M block data, 417 denotes a pixel M × M block data embedding unit, and 418 denotes a restored multivalued image.

A restored image generation / interpolation filtering method for generating a multivalued image from a dither image will be described with reference to the block diagram shown in FIG. First, the entire dither image 41 is divided into sub-blocks of an M × M pixel matrix (process 42). The obtained block data 43 is 2D-D
It is a unit for performing spectrum conversion in the CT processing 44 unit, and sequentially has a luminance value of M × M matrix size (black:
0, white: 255) for 2D-DCT processing
Find a spectrum value in the T region. Thus, dither spectrum M × M block data 45 is obtained. Next, in order to convert the dither image into a multi-valued image, a dither spectrum and a multi-value filter (having a low-pass filter characteristic for suppressing high-frequency components unique to the dither image)
(A value obtained by multiplying spectral values on the same spatial frequency is referred to as a multivalued spectrum). Further, in the multi-valued spectrum M × M block data 48, the spectrum M × M block data value suppressed by the multi-valued filter is close to the value 0, which is a wasteful information amount. Therefore, the domain having a value of 0 in the decimation processing unit 49 is truncated to obtain N × N block decimation spectrum N × N block data 410.

On the other hand, the spectrum M × M is obtained from the interpolation coefficient output from the interpolation coefficient analysis section 411 for restoring the high-frequency component cut off by the decimation processing section 49 and the decimation spectrum N × N block data.
Generate and interpolate data.

FIG. 2 illustrates the mode of processing of the interpolation coefficient analysis unit.
FIG. In FIG. 2, the decimation spec
Of the N × N block data, spectrum n × n blocks
Data (however, n = N × N / M, n: integer
Focus) on each element of the spectrum n × n block data.
Element b_ijFrom each element a of the spectrum N × N block data
_lmIs configured. In the interpolation coefficient analysis unit, the image filter shown in FIG.
In the Taling flow, for all image blocks in advance
Find N × N block data of decimation spectrum
A_lmIs b_ijBased on equation (1), which is a linear combination sum of
The primary term (k on the left side of equation (2)_lmij) And a constant term (S_ij)
It is determined by the least squares approximation method.

Next, the spectrum interpolation processing section 413 will be described. FIG. 3 is a diagram for explaining a mode of processing of the spectrum interpolation processing unit. In FIG. 3, the spectrum N
Each time spectrum N × N block data is generated by multiplying each spectrum n × n block data obtained by dividing × N block data into regions 1 to 4 by an interpolation coefficient output from the interpolation coefficient analysis unit, the spectrum N × N block data is generated. The spectrum data corresponding to the spectrum M × M block data is obtained in the order of the area corresponding to the × N block data.

The restored spectrum M × M block data 416 is obtained by subjecting the restored spectrum M × M block data 414 to spectrum inverse transform (2D-IDCT).

In the above image filtering processing, a sub-block image of M × M matrix size is cut out from the original image, and image filtering is sequentially performed, and finally, a restored multi-valued image 418 can be generated.

[Embodiment 2] FIG. 4 shows another embodiment of the present invention. 4, reference numeral 71 denotes a dither image, 72 denotes a pixel M × M block data extraction unit, 73 denotes a pixel M × M block data, 74 denotes an M × M block 2D-DCT processing unit, and 75 denotes a dither spectrum M × M block data. , 76 is a multi-valued filter M × M block data, 77 is a multi-valued image filtering process, 78 is a multi-valued spectrum M × M block data, 79 is a decimation processing unit, 710 is a decimation spectrum N × N block data, 711 is an interpolation coefficient analysis unit, 712 is an interpolation coefficient for interpolating spectrum M × M block data from spectrum N × N block data, 713 is a spectrum interpolation processing unit, 714 is restored spectrum M × M block data, and 715 is M × M block data. M
The block 2D-IDCT processing unit 716 is a pixel M ×
M block data, 717 denotes a pixel M × M block data embedding unit, 718 denotes a restored multi-valued image, 719 denotes a smoothness reproduction filter M × M block data, and 720 denotes a correction filtering processing unit for reproducing smoothness. .

A restoration image correction filtering method using a correction filter for reproducing smoothness will be described with reference to a processing block diagram shown in FIG. Reference numeral 7 in FIG.
1 to 718 correspond to the reference numerals 41 to 418 shown in FIG. 1, the description thereof is omitted, and the M × M block size interpolation spectrum output from the spectrum interpolation processing unit 713 is converted to a 2D-IDCT processing unit 715. Before inputting to the, only the part to be subjected to image filtering by the correction filter is different, and only the processing part of this difference will be described.

In FIG. 4, a spectrum interpolation processing section 71
In step 3, the interpolated spectrum M × M block data is converted into a smoothness reproduction filter M × M block data 719.
Is performed by filtering.

FIG. 5 shows an aspect of the correction filtering process. In each of the spatial frequency u direction (horizontal direction in the drawing) and the spatial frequency v direction (vertical direction in the drawing), a frequency ω _p corresponding to a level of −3 dB and a frequency ω _s corresponding to a level of −20 dB are set and illustrated. Histograms h (u) and h obtained smoothness corresponding to equations (3) and (4)
(V).

As an example of the smoothness filter characteristic used in the correction filtering process, the filter has a characteristic that slightly suppresses a high frequency component. A pass band up to N × N blocks is a pass band, and other high frequency components are a stop band. , Butt
It has an erworth type filter characteristic. The amplitude value of the DC component is set to 1, and a low-pass filter is formed up to the high-frequency component with characteristics according to Expressions (3) and (4). In the two-dimensional image filtering, first, the low-pass filter characteristic of Expression (3) is calculated in the row direction (u direction), and then the low-pass filter characteristic of Expression (4) is calculated in the column direction (v direction). Is realized.

Third Embodiment FIG. 6 is a block diagram showing a method for implementing the third embodiment. In the figure, reference numeral 91 denotes a color dither image, 92 denotes a pixel M × M block data extraction unit, 93 denotes pixel M × M block data, and 94 denotes M × M block data.
M block 2D-DCT processing section, 95 is dither spectrum M × M block data, 96 is multi-valued filter M × M
Block data 97, multi-level image filtering processing, 98 multi-level spectrum M × M block data, 9
9 is a decimation processing unit, 910 is a decimation spectrum N × N block data, 911 is an interpolation coefficient analysis unit, 912 is an interpolation coefficient for interpolating spectrum M × M block data from a spectrum N × N block, 913 is a spectrum interpolation processing unit, 914 is the restored spectrum M ×
M block data, 915 is M × M block 2D-ID
CT processing unit 916, pixel M × M block data,
Reference numeral 917 denotes a pixel M × M block data embedding unit;
18 is a color restoration multi-valued image, 919 is a smoothness reproduction filter M × M block data, 920 is a correction filtering processing unit for reproducing smoothness, 921 is a color separation processing unit, and 922 is a color synthesis processing unit. .

The reference numerals 91 to 920 in FIG.
Substantially correspond to the reference numerals 71 to 720 in FIG. That is, as compared with the block diagram of FIG. 4, a color separation processing unit 921 and a color synthesis processing unit 922 are newly added, and the input image data is a color dither image (a 3-bit image in which each RGB color is represented by 1 bit). Yes, the output is a color image (R
The color image restoration process is performed as a 24 bit image in which each of the GB colors is expressed by 8 bits. The image filtering process in the block diagram is described in FIG.
Although the description is omitted, image filtering is performed using a multi-value filter corresponding to each color and a correction coefficient (obtained by performing correction analysis).

In order to separate the input color dither image into three colors, the color separation processing section 921 performs a separation process from the 3-bit color dither image into three 1-bit color images. The 1-bit color image to be color-separated is processed in the order of RGB (interleave processing) or after the entire monochrome image is processed (non-interleave processing), and then the next entire monochrome image is processed. A color image restoration process is performed for each single color. Since the multi-valued filter, the interpolation coefficient analyzer, the spectrum interpolation, and the like are processed for each color, each data output as a correction coefficient may be different depending on the corresponding color characteristics. The color synthesis processing unit 922 is a processing unit for synthesizing images for each color separated on the input side, and generates a color restoration image processed for each color.

[0031]

As described above, according to the present invention,
By using a multi-value generation filter and a restoration / correction filter in a spectral domain for a deteriorated image such as a low-bit image without assuming an original image and a noise model in a spectral domain, the original image is approximated. It is possible to generate a restored image to be reproduced at the same time.

[Brief description of the drawings]

FIG. 1 shows a block diagram of image filtering according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a mode of processing of an interpolation coefficient analysis unit.

FIG. 3 is a diagram illustrating a mode of processing of a spectrum interpolation processing unit.

FIG. 4 shows another embodiment of the present invention.

FIG. 5 shows an aspect of a correction filtering process.

FIG. 6 is a block diagram illustrating a method for implementing a third embodiment;

FIG. 7 shows a configuration example of a conventional image restoration.

FIG. 8 illustrates an example of multi-level image filtering.

FIG. 9 is a diagram illustrating a decimation process.

[Explanation of symbols]

41 Dither image 42 Pixel M × M block data extraction unit 43 Pixel M × M block data 44 M × M block 2D-DCT processing unit 45 Dither spectrum M × M block data 46 Multi-value filter M × M block data 47 Multi-value Image filtering processing 48 Multi-valued spectrum M × M block data 49 Decimation processing unit 410 Decimation spectrum N × N block data 411 Interpolation coefficient analysis unit 412 Interpolation coefficient for interpolating spectrum M × M block data from spectrum N × N block data 413 Spectrum interpolation processing unit 414 Reconstructed spectrum M × M block data 415 M × M block 2D-IDCT processing unit 416 pixel M × M block data 417 pixel M × M block data embedding unit 418 Value image 71 Dither image 72 Pixel M × M block data extraction unit 73 Pixel M × M block data 74 M × M block 2D-DCT processing unit 75 Dither spectrum M × M block data 76 Multilevel filter M × M block data 77 Multi-valued image filtering processing 78 Multi-valued spectrum M × M block data 79 Decimation processing unit 710 Decimation spectrum N × N block data 711 Interpolation coefficient analysis unit 712 Interpolates spectrum M × M block data from spectrum N × N block data Interpolation coefficient 713 Spectrum interpolation processing unit 714 Reconstructed spectrum M × M block data 715 M × M block 2D-IDCT processing unit 716 Pixel M × M block data 717 Pixel M × M block data embedding unit 718 Restored multi-valued image 719 Smoothness reproduction filter M × M block data 720 Correction filtering processing unit for smoothness reproduction 91 Color dither image 92 Pixel M × M block data extraction unit 93 Pixel M × M block data 94 M × M block 2D-DCT processing section 95 Dither spectrum M × M block data 96 Multi-level quantization filter M × M block data 97 Multi-level quantization image filtering processing 98 Multi-level quantization spectrum M × M block data 99 Decimation processing section 910 Decimation spectrum N × N Block data 911 Interpolation coefficient analysis unit 912 Interpolation coefficient for interpolating spectrum M × M block data from spectrum N × N block data 913 Spectrum interpolation processing unit 914 Reconstructed spectrum M × M block data 915 M × M block 2D-IDCT processing unit 916 Pixel M × M block data 917 Pixel M × M block data embedding unit 918 Color restoration multi-valued image 919 Smoothness reproduction filter M × M block data 920 Correction filtering processing unit for smoothness reproduction 921 Color separation processing unit 922 Color synthesis processing unit 11 Original image spectrum 12 Image quality deteriorated image spectrum 13 Restored image spectrum 14 Noise analysis processing unit 15 Noise pattern model 16 Noise inverse characteristic data 17 Noise addition processing 18 Noise removal processing 21 Binary image ( Low-bit image such as a dither image) 22 pixel M × M block data 23 spectrum conversion processing 24 spectrum M × M block data 25 multi-level filter M × M block data 26 multi-level image filtering processing 27 decimation processing 28 Decimation spectrum N × N block data 29 Spectral inversion processing 210 Pixel N × N block data 211 Multi-valued image

Claims (3)

[Claims] 1. A low bit image is converted into an original high bit high quality multivalued image or a continuous tone image by M × M (M: integer).
In the image filtering processing for restoring in units of blocks, multi-level quantization having amplitude characteristics for suppressing high-frequency components of the block data is performed on spectrum M × M block data obtained by spectral conversion of a pixel area M × M block cut out from a low bit image. The image is filtered using a filter, and the spectrum N × is obtained by a decimation process for discarding the suppressed high-frequency components in an area other than N × N blocks (N: integer, 0 <N <M) in the spectrum M × M data.
The data is processed into data for only N blocks, and the spectrum n × n (n
= N × N / M) from each element b _{ij of the} block to the spectrum N
An interpolation coefficient for predicting each element a _ij of the × N block is obtained, and spectrum M × M block data is generated from the interpolation coefficient and each element of the spectrum N × N block. A restored image interpolation filtering method characterized by restoring an image. 2. The method according to claim 1, wherein the spectrum M × M block data obtained by spectrum interpolation using the spectrum N × N block data and the correction coefficient is converted into a pixel area by a spectrum inverse transform. Spectral response function h (x, y) (0 ≦ x <M, 0 ≦ y) designed to generate smooth image shading in space
A restored image correction filtering method characterized by performing image filtering using <M) to reproduce smoothness of image density. 3. The color image processing method according to claim 1, wherein when the original image is a color image, the color component images are separated by a color separation process, and multi-valued spectrum N × N block data is generated for each color component image. The spectrum interpolation processing is performed from the spectrum N × N block data to the spectrum M × M block data by using the interpolation coefficient corresponding to, and the color image is re-synthesized by the color synthesis processing to restore the original color image. Color restoration image interpolation and correction filtering method.