JP2009026032A - Resolution conversion device and resolution conversion method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce throughput and processing time in the case of enlarging an image by fractal conversion, and to obtain an enlarged image with the characteristics of one or both of a fractal conversion enlarged image and an enlarged image enlarged by another enlargement method such as linear interpolation. <P>SOLUTION: A scaling part 2 enlarges an input image 1 by using an enlargement algorithm such as linear interpolation other than fractal conversion. An iterative function system fractal decoding part 6 operates fractal decoding to fractal encoded data based on an enlargement rate by iteratively repeating decoding processing to an image 3 enlarged by the scaling part 2 as a decoded initial image. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a resolution conversion apparatus and a resolution conversion method for converting an input image into a high-resolution image having a larger number of pixels than the image.

In general, it is known that a natural image has a strong correlation between neighboring pixels. In the case of resolution conversion for enlarging an image, an interpolated pixel at the time of enlarging can be generated by filtering the neighboring pixels using the correlation.
In addition, as a technique proposed by Jacquin et al. For resolution conversion using fractal self-similarity of images as a technique for obtaining a good decoded image, the input image is divided into a plurality of range blocks, and this range block A domain block whose size is enlarged based on the enlargement ratio is set. A window corresponding to a domain block is scanned in the input image, and a block having the highest self-similarity is searched from the image. When domain blocks having high self-similarity with each range block are obtained from the input image and all range blocks are replaced with domain blocks, an image with a predetermined enlargement ratio can be obtained.

However, when the image is enlarged by generating the interpolation pixel by the above-described filter processing, image deterioration such as blurring or jaggy is likely to occur.
Also, in the case of image enlargement by replacement using fractal self-similarity, high-fidelity conversion is possible within a block, but each domain block is extracted from an arbitrary place in the input image, so that each other There is no continuity at the boundary. Therefore, an enlarged image obtained by simple replacement causes discontinuity at the block boundary, resulting in degradation of image quality.

In order to solve such a problem of the image enlargement method, the following Patent Document 1 discloses that an interpolated enlarged image and a fractal transformed enlarged image obtained by linear interpolation according to the enlargement magnification are generated, and then both enlarged images are displayed. Compared to,
(1) A method of adopting a linear interpolation enlarged image when the difference is large, and a fractal transformation enlarged image in other cases,
(2) A method of adding the product of the difference and the weighting coefficient to the linear interpolation enlarged image is disclosed.
JP-A-11-331595 (Claims 1 and 2)

  However, in the conventional image enlargement technique described above, after generating two types of enlarged images, that is, the linearly-interpolated enlarged image and the fractal transformation enlarged image, the enlarged images of the two are compared and various processes such as product-sum operation are performed. As a result, the amount of processing increases and the processing time increases.

  The present invention has been made in view of the above-described problems of the prior art, and the processing amount and processing time are reduced when an enlarged image enlarged by another enlargement method other than the fractal conversion method is corrected using the fractal conversion method. Resolution conversion apparatus and resolution conversion method capable of obtaining a magnified image that can be reduced, and can be obtained by maintaining either a fractal transformed enlarged image and an enlarged image enlarged by another method such as linear interpolation, or both features The purpose is to provide.

In order to achieve the above object, the present invention provides a resolution converting apparatus that converts an input image into a high-resolution image having a larger number of pixels than the image, and the input image is subjected to a predetermined enlargement technique. Then, in accordance with the set enlargement ratio, enlargement means (2) that enlarges the input image to a high-resolution enlarged image having a larger number of pixels than the input image, and fractal encoded data that includes a parameter relating to the image size. Fractal encoding means (4) for performing fractal encoding, parameter conversion means (5) for converting a parameter relating to the image size, which is fractal encoded by the fractal encoding means, in accordance with the enlargement ratio, and the enlargement The fractal encoded data encoded by the fractal encoding means, using the enlarged image enlarged by the means as a decoded initial image, Fractal decoding is performed based on the converted parameters relating to the image size, and the enlargement is performed by repeating a decoding process using the decoded image obtained by fractal decoding as a decoding initial image of the next fractal decoding as many times as desired. And a fractal decoding means (6) for correcting the enlarged image enlarged by the means.
According to a second aspect of the present invention, there is provided a resolution conversion method for converting an input image into a high-resolution image having a larger number of pixels than that image, and the input image is subjected to a predetermined enlargement method according to a set enlargement ratio. , A step of enlarging to a high resolution enlarged image having a larger number of pixels than the input image, a step of fractal encoding the input image into fractal encoded data including a parameter relating to an image size, and the step of fractal encoding A step of converting the parameter relating to the image size, which has been fractal encoded by the step of, according to the enlargement ratio, and the step of performing the fractal encoding using the enlarged image enlarged by the enlargement step as a decoded initial image. The fractal encoded data is converted into a parameter relating to the converted image size. The fractal decoding is performed on the basis of the fractal decoding, and the decoded image obtained by performing the fractal decoding is used as a decoding initial image of the next fractal decoding. A step to fix,
A resolution conversion method is provided.

According to the resolution conversion apparatus and the resolution conversion method of the present invention, fractal decoding is performed using an enlarged image enlarged by another enlargement technique (algorithm) such as linear interpolation as a decoded initial image, and the enlarged image is corrected. Since the process after conversion is not necessary and the convergence of decoding is accelerated, the amount of processing and the processing time can be reduced.
In addition, since fractal transformation is an iterative process, if the process is stopped before the iterative process is converged, the features of the magnified image enlarged by other enlargement methods such as linear interpolation remain, and the process can be repeated until the iterative process is converged. For example, since only the feature of the fractal transformation enlarged image remains, an enlarged image in which either the fractal transformation enlarged image and the enlarged image enlarged by another enlargement method such as linear interpolation, or the features of both can be obtained.

Embodiments of the present invention will be described below with reference to the drawings.
The resolution conversion apparatus and the resolution conversion method according to the present invention perform fractal decoding using an enlarged image enlarged by another enlargement method such as linear interpolation as a decoding initial image, and correct the enlarged image, thereby processing amount and processing time. In addition, it is possible to obtain a magnified image in which either the fractal-transform magnified image, the magnified image magnified by another magnification technique such as linear interpolation, or the characteristics of both are left.
Before describing the resolution conversion apparatus according to the present invention, first, a fractal encoding / decoding algorithm using a general iterative function system as a prerequisite technique will be described with reference to FIG.

<Fractal coding algorithm>
First, the fractal encoding algorithm will be described.
In the fractal encoding process, an input image f (x, y) (0 ≦ x ≦ Xmax, 0 ≦ y ≦ Ymax) as shown in FIGS. 1A and 1B is shown in FIG. As shown in FIG. 1E, m′max × n′max pixels (mmax <m) larger than that corresponding to each range block Rij are divided. This is done by searching for a domain block Dki with 'max, nmax <n'max) to find a transformation λij (·) that gives an optimal approximation. This transformation λij (·) is described as a reduced affine transformation.

When the range block Rij is flat, since the block Rij can be sufficiently approximated only by the average value Rave, it is determined as a shade block Sij. At this time, the domain block Dki is not searched, and only the average value Rave of the range block Rij and the block discrimination bit τij are used as codes. Thereby, the bit rate and the search time can be shortened.
In order to discriminate the shade block Sij, the variance σ ² of the range block Rij is used, and the range block Rij that is equal to or less than a preset threshold value is set as the shade block Sij. A range block Rij in which the variance σ ² of the range block Rij is not less than or equal to the threshold is determined as an edge block Eij.
For the edge block Eij, the domain block Dki and the optimum transformation λij (·) are searched brute force.

Here, conversion performed on the range block Rij shown in FIG. 1C and the domain block Dki shown in FIG. 1E will be described below.
First, a DC component (average value Rave) is separated from the range block Rij, and a range block R′ij from which the average value Rave is separated is created.
R′ij (m, n) = Rij (m, n) −Rave (1)

Next, the domain block Dki is extracted, and the reduction conversion s (•) is performed by arithmetically averaging the pixel values corresponding to the extracted domain block Dki. Further, as with the range block Rij, the DC component (average value Dave) is separated, and isometric transformation ε (·) is performed to generate the separated and transformed domain block D′ ki as shown in FIG. To do.
D′ ki (m, n) = ε (s (Dki (m ′, n ′)) − Dave) (2)

As shown in FIG. 2, the following eight types of isometric transformations ε (•) are used.
1) Identity transformation 2) Rotate 90 degrees clockwise 3) Rotate 180 degrees clockwise 4) Rotate 270 degrees clockwise 5) Reflection on diagonal 45 of +45 degrees 6) Reflection on diagonal 2 of 45 degrees 7) Reflection on the horizontal axis 8) Reflection on the vertical axis

Finally, as shown in FIG. 1D, gray level scaling α is applied to the separated and converted domain block D′ ki.
R′ij (m, n) ≈α · D′ ki (m, n) (3)
At this time, α is calculated from the least square method such that the relationship of the above equation (3) is satisfied. Α = (ΣR′ij (m, n) · D′ ki (m, n)) / (Σ (D′ ki (m , n)) ² ) ... (4)
It is calculated by.

The above conversion processing operation is repeated for the domain block Dki and conversion parameters in a round-robin manner so that the mean square error between the range block R′ij from which the average value Rave is separated and the domain block Dki is the lowest. A domain block Dki and a transformation λ giving an optimum approximation are obtained.
For the range block Rij, the block discrimination bit τij, the average value Rave of the range block Rij, the position (k, l) of the domain block Dki, the isometric transformation ε, and the scaling coefficient α are finally encoded as conversion parameters. The above is the fractal encoding algorithm.

<Fractal decoding algorithm>
Next, the fractal decoding algorithm will be described.
The fractal decoding process is realized by repeatedly converting the domain block Dki to the range block Rij using an arbitrary initial image. The procedure is described below.
First, an image output area having the same size as that of the input image f (x, y) is prepared, and is divided into range blocks Rij to obtain a decoded image output F (x, y).
Further, an arbitrary initial image finit (x, y) (0 ≦ x ≦ Xmax, 0 ≦ y ≦ Ymax) is prepared, and subsequently, a domain block Dki corresponding to each range block Rij is extracted, and this domain block Dki is extracted. The range block Rij is replaced by the conversion represented by the following equation (5).
Rij = α · ε (s (Dki (m ′, n ′)) − Dave) + Rave (5)
However, Rij = Rave is replaced for the shade block Sij. The above processing is executed for all range blocks Rij, and the first decoding processing is performed. Further, by repeating the same conversion using the decoded image as an initial image, the gray level of the image converges and a reproduced image is obtained. The above is the fractal decoding algorithm.

<Image enlargement processing>
A method of enlarging the input image f (x, y) at the enlargement ratio A using the above fractal encoding algorithm will be described with reference to FIG.
First, the input image f (x, y) is encoded using the above-described fractal encoding algorithm, and the block discrimination bit τij, the average value Rave of the range block Rij, and the position ( k, l), isometric transformation ε, and scaling factor α.
Next, before entering the fractal decoding process, the initial image finit (x, y) shown in FIG. 3A and the decoded image output F (x, y) shown in FIG. ) And (d), both the vertical and horizontal directions are set to an enlarged size according to the enlargement ratio A.
Initial image f'init (x, y),
Decoded image output F '(x, y)
0 ≦ x ≦ (Xmax · A), 0 ≦ y ≦ (Ymax · A)

Accordingly, the size of the range block Rij and the domain block Dki is also set to an enlarged size in accordance with the enlargement ratio A in both the vertical and horizontal directions. Further, the position (k, l) of the domain block Dki is also converted at the enlargement ratio A as shown by the following expression (6).
k ′ = k · A
l '= l · A
0 ≦ k ′ ≦ (Xmax · A), 0 ≦ l ′ ≦ (Ymax · A) (6)

The meaning of the information encoded by fractal encoding encodes the relationship that a certain part of the image is similar to other parts of the image. Since the similar relationship is maintained even when the image is enlarged, only the parameters related to the image size, such as the initial image, the decoded image, each block size, and the position information, are decoded at a predetermined magnification. It is possible to obtain a magnified image by enlarging and performing the previous decoding process.
The initial image may be an arbitrary image, but by using an enlarged image created by another algorithm as an initial image, the convergence of decoding is accelerated, and the enlarged image can be used for correction and correction processing of the enlarged image by another algorithm.

<Embodiment of the present invention>
An example of the hardware configuration of the embodiment of the resolution conversion apparatus of the present invention is shown in FIG.
In FIG. 4, an input image 1 that is f (x, y) of Xmax × Ymax pixels is a scaling unit 2 to obtain a decoded initial image Finit (x, y) of Xmax · A × Ymax · A pixels. Sent to.
The scaling unit 2 enlarges the input image 1 with the set magnification A, and obtains a fractal decoded initial image 3 (enlarged image) that is Finit (x, y).
This enlargement method (algorithm) may be any enlargement algorithm such as linear interpolation. Therefore, when another enlargement algorithm is used, this resolution conversion apparatus can be a correction / correction processing apparatus for an enlarged image using another enlargement algorithm.

On the other hand, simultaneously with the above-described processing, the input image 1 is sent to the iterative function system fractal encoding unit 4 and subjected to fractal encoding. As a result, the block discrimination bit τij, the average value Rave of the range block Rij, Coding parameters such as the position (k, l) of the domain block Dki, isometric transformation ε, and scaling coefficient α are obtained. Of these encoding parameters, only the parameters related to the image size are sent to the encoding parameter conversion unit 5 and converted at the set magnification A. Other parameters than the parameters related to the image size sent to the encoding parameter conversion unit 5 are sent to the iterative function system fractal decoding unit 6 as they are.
In the case of the fractal encoding of this example, the parameter converted by the encoding parameter converting unit 5 is the position vector (k, l) of the domain block Dki, and the conversion is performed using the above-described conversion equation (6). Then, the converted position vector (k ′, l ′) of the domain block Dki is sent to the iterative function system fractal decoding unit 6.

The iterative function system fractal decoding unit 6 uses the image 3 enlarged by the scaling unit 2 as a decoded initial image Finit (x, y), and the code sent from the iterative function system fractal encoding unit 4 and the encoding parameter conversion unit 5. Decoding is performed using the optimization parameter. At this time, the image size and block size are enlarged at a magnification A.
This fractal decoding process is executed for all the range blocks Rij obtained by dividing the input image 1, and after the first decoding process, the same decoding process is repeated with the decoded image as an initial image. Repeated. By repeating such decoding processing repeatedly, the gray level of the image converges, and a decoded image 7 is obtained and output.
The decoded image 7 output by the above configuration becomes an enlarged image F (x, y) A times as large as the input image 1. In the case of iterative function system fractal decoding, the decoding process is repeated until the image converges. However, when the decoding process is terminated at a certain number of times, the characteristics of the decoded initial image 3 (ie, enlargement by another enlargement method such as linear interpolation) is performed. The resolution conversion apparatus becomes a correction / correction processing apparatus for the enlargement algorithm of the scaling unit 2 as a result.
Further, when the decoding process is repeated until the image is completely converged, it is not affected by the initial decoding image 3, so that it becomes an enlarged image under the influence of the fractal encoding alone, and only the features of the fractal transformed enlarged image remain.

It is explanatory drawing which shows the self-similar domain block search process in fractal encoding. It is explanatory drawing which shows the equal length conversion process in fractal encoding. It is explanatory drawing which shows the image expansion process in fractal encoding. It is a block diagram which shows the resolution conversion apparatus of one embodiment of this invention.

Explanation of symbols

1 Input image 2 Scaling unit 3 Decoded initial image (enlarged image)
4 Iterative Function System Fractal Encoding Unit 5 Coding Parameter Conversion Unit 6 Iterative Function System Fractal Decoding Unit 7 Decoded Image (Enlarged Image)

Claims (2)

In a resolution conversion device that converts an input image into a high-resolution image having more pixels than the image,
Enlarging means for enlarging the input image into a high-resolution enlarged image having a larger number of pixels than the input image according to a set enlargement ratio by a predetermined enlargement method;
Fractal encoding means for fractal encoding the input image into fractal encoded data including parameters relating to the image size;
Parameter conversion means for converting a parameter relating to the image size, which is fractal encoded by the fractal encoding means, in accordance with the enlargement ratio;
Using the enlarged image enlarged by the enlargement means as a decoded initial image, the fractal encoded data encoded by the fractal encoding means is fractal-decoded based on the converted parameters relating to the image size, and the fractal Fractal decoding means for correcting the enlarged image enlarged by the enlargement means by repeating a decoding process using the decoded image obtained by decoding as a decoding initial image of the next fractal decoding, an arbitrary number of times;
A resolution conversion device. In a resolution conversion method for converting an input image into a high-resolution image having more pixels than the image,
Enlarging the input image into a high-resolution enlarged image having a larger number of pixels than the input image according to a set enlargement ratio by a predetermined enlargement method;
Fractal encoding the input image into fractal encoded data including parameters relating to image size;
Converting the parameter relating to the image size, which is fractal encoded by the fractal encoding step, according to the enlargement ratio;
The fractal encoded data encoded by the fractal encoding step is subjected to fractal decoding based on the converted parameter relating to the image size, using the enlarged image expanded by the expanding step as a decoded initial image. Correcting the enlarged image enlarged by the enlargement step by repeating a decoding process using the decoded image obtained by the fractal decoding as a decoding initial image of the next fractal decoding, an arbitrary number of times;
A resolution conversion method.

Images