KR101024282B1 - Apparatus and method for image reproduction in compressed domain - Google Patents

Abstract

PURPOSE: An image decoder and a method thereof are provided to enhance brightness information and detail information of an original image by distributing a DCT block and to eliminate block phenomenon. CONSTITUTION: A block decomposing unit(120) decomposes each DCT block by a preset conversion matrix. The block decomposing unit generates sub blocks having the frequency distribution of conversion coefficients. A brightness information enhancement unit(130) improves the brightness information in a shadow region and a bright region of the original image. A detail information enhancement unit(140) enhance detail information of the original information.

Description

Apparatus and method for image reproduction in compressed domain

The present invention relates to an apparatus and method for image restoration in a compressed region, and more particularly, to an apparatus and method for reconstructing a compressed image by processing transform coefficients of an image compressed by discrete cosine transform.

The advent of the Internet and multimedia has made it possible for information to be compressed and delivered in a limited band of networks. Among the transformation techniques for data compression, discrete cosine transform is the most widely used method because of its excellent energy concentration and minimal correlation. DCT is used in standards for the compression of still images and video sequences such as JPEG and MPEG.

Image enhancement in the compressed region, ie dynamic range compression and contrast enhancement, is performed directly on the transform coefficients of the image compressed by the DCT. Image quality improvement methods in such a compressed region include an α-rooting technique, a modified unsharp masking technique, and a multi-scale contrast measure technique.

Such compressed region processing methods have advantages of low computational complexity and easy observation and adjustment of frequency synthesis of an image. However, these techniques can lead to block artifacts such as unnecessary edges at the block boundary, and it is reported that all parts of the image cannot be improved at the same time.

Therefore, there is a need for a method for effectively restoring an image by suppressing an artificial phenomenon such as a block phenomenon in the DCT compression region and improving the quality of the compressed image.

The technical problem to be achieved by the present invention, image reconstruction apparatus and method for improving the brightness information and detail information of the original image by modifying the value of the transform coefficient constituting a plurality of DCT blocks generated by the discrete cosine transform on the original image To provide.

Another technical problem to be solved by the present invention is an image restoration method for improving brightness information and detail information of an original image by modifying values of transform coefficients constituting a plurality of DCT blocks generated by discrete cosine transform on the original image. The present invention provides a computer-readable recording medium that records a program for execution on a computer.

In order to achieve the above technical problem, the apparatus for reconstructing an image in the compressed region according to the present invention includes a plurality of generated original images divided into a plurality of unit blocks and discrete cosine transform and quantization transform are performed on each unit block. A compressed image input unit receiving a DCT block; A block decomposition unit for generating a plurality of sub-blocks having a frequency distribution equal to a frequency distribution of transform coefficients constituting the DCT block by decomposing the DCT block by a predetermined transform matrix; Brightness information for improving the brightness information in the dark region and the bright region of the original image by modifying the value of the transform coefficient corresponding to the DC component among the transform coefficients constituting the sub-block by the first transform function which is a nonlinear function Improvement unit; A detail information improving unit for modifying a value of a transform coefficient corresponding to an AC component among the transform coefficients constituting the sub-block by a second transform function which is a nonlinear function to improve detail information included in the original image; And a block reconstruction unit configured to reconstruct a transformed DCT block by applying an inverse matrix of the transform matrix to subblocks formed of transformed transform coefficients when values of the transform coefficients are transformed for all subblocks generated from the DCT block. It includes;

In order to achieve the above technical problem, an image reconstruction method in a compressed region according to the present invention includes: (a) an original image is divided into a plurality of unit blocks, and discrete cosine transform and quantization transform are performed on each unit block. Receiving a plurality of generated DCT blocks; (b) generating a plurality of subblocks having a frequency distribution equal to a frequency distribution of transform coefficients constituting the DCT block by decomposing the DCT block by a predetermined transform matrix; (c) transforming a value of a transform coefficient corresponding to a DC component among the transform coefficients constituting the sub-block by a first transform function that is a nonlinear function to improve brightness information in a dark region and a bright region of the original image; step; (d) modifying a value of a transform coefficient corresponding to an AC component among transform coefficients constituting the sub-block by a second transform function which is a nonlinear function to improve detail information included in the original image; And (e) reconstructing a transformed DCT block by applying an inverse matrix of the transform matrix to subblocks formed of transformed transform coefficients when values of the transform coefficients are transformed for all subblocks generated from the DCT block. Step;

According to the apparatus and method for image reconstruction in a compressed region according to the present invention, a ringing phenomenon that may occur when the original image is processed in the compressed region by modifying a value of a transform coefficient constituting a subblock generated by decomposing a DCT block. And block phenomenon can be eliminated. In addition, by transforming the values of the transform coefficients corresponding to the DC component and the transform coefficient corresponding to the AC component among the transform coefficients constituting the sub-block in a separate method, it is possible to effectively improve the brightness information and the detail information of the original image.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the image restoration apparatus and method in a compressed region according to the present invention.

1 is a block diagram showing the configuration of a preferred embodiment of an image restoration apparatus in a compressed region according to the present invention.

Referring to FIG. 1, the apparatus for reconstructing an image according to the present invention includes a compressed image input unit 110, a block decomposition unit 120, a brightness information improving unit 130, a detail information improving unit 140, and a block reconstructing unit 150. It is provided.

The compressed image input unit 110 receives a plurality of DCT blocks generated by dividing an original image into a plurality of unit blocks and performing discrete cosine transform and quantization transform on each unit block.

In DCT-based image compression, the original image is first divided into a series of unit blocks having a size of 8x8 pixels. Each unit block is then transformed into a DCT block in the spatial frequency domain by a forward DCT. The DCT block transformed into the spatial frequency domain is composed of 8 × 8 DCT transform coefficients corresponding to the pixels constituting the unit block.

The values of the DCT transform coefficients constituting one DCT block are obtained from each pixel of the unit block by a transform equation represented by Equation 1 below.

Here, X (u, v) is the value of the transform coefficient corresponding to the (u, v) position in the DCT block, and x (i, j) is the (i, v) position corresponding to the (u, v) position of the DCT block in the unit block. , j) is a pixel value at the position, and the c (·) value is determined as follows.

In one DCT block of size 8 × 8, X (0,0) is a transform coefficient corresponding to a DC component, and the rest of 63 are transform coefficients corresponding to an AC component. FIG. 2 shows 64 DCT basic functions. These basic functions are arranged in the order of increasing spatial frequency from the upper left to the lower right, and the conversion coefficients at positions where straight lines connected to each other by the same numbers shown in FIG. 2 are considered to belong to the same frequency band. Therefore, 64 transform coefficients constituting one DCT block may be classified into 14 frequency bands. In Equation 1, X (u, v) represents the contribution from the basic function located in the u th row v th column among the basic functions of FIG. 2.

After DCT is performed, 64 transform coefficients of each DCT block are quantized, aligned in a zigzag scan direction, and then entropy coded. This process is performed for all DCT blocks to obtain a compressed image. Entropy coding is a run-length coding method in which a long string of zeros is effectively compressed. Entropy coding for each quantized DCT block leaves only a small number of nonzero transform coefficients in each block.

In order to restore the original image by the decoding process, the above-described image compression process is reversed. That is, the compressed image is first entropy decoded and inversely quantized by the quantization parameter to obtain a plurality of DCT blocks. For each DCT block, an inverse DCT is performed to obtain a unit block. The inverse DCT can be expressed as Equation 2 below.

Where X (u, v) corresponds to the value of the transform coefficient corresponding to the (u, v) position in the DCT block, and x (i, j) corresponds to the (u, v) position of the DCT block in the restored unit block Pixel value at position (i, j).

The compressed image input unit 110 of the image reconstruction apparatus according to the present invention performs the following reconstruction process on the compressed region of the original image by receiving a DCT block before performing inverse quantization and inverse DCT.

The block decomposition unit 120 decomposes the DCT block by using a preset transformation matrix to generate a plurality of subblocks having the same frequency distribution as that of the transform coefficients constituting the DCT block.

In order to explain the spatial relationship between DCT blocks in one dimension, {x (i)} (i = 0, 1, ..., MN-1) is referred to as a sequence having a length MN. This sequence may be divided into M blocks (or subsequences) each containing N data. In the block DCT space, an N-point DCT may be applied to each block. Therefore, the N-point DCT for the p-th DCT block is expressed by Equation 3 below.

Here, C _p (u) is a transform coefficient obtained by the N-point DCT of the u-th data of the p-th block among the M blocks, N is the number of data contained in the p-th block, c (u) is mathematics The value is determined in the same manner as in Equation 1.

On the other hand, the MN-point DCT for the sequence x (i) is expressed by the following equation (4).

Here, C (u) is a conversion coefficient obtained by the MN-point DCT for the u-th data among the MN pieces of data.

The block DCT transform represented by Equation 3 is an orthonormal extension of the sequence {x (i)} by the set of MN basis vectors, and the MN basis vectors are each obtained from the basis vectors of the N-point DCTs. Lose. Therefore, M blocks composed of N-point DCT transform coefficients may be converted into MN-point DCT transform coefficients by an invertible linear transformation. In other words, MN × MN for the composite DCT C ^{( MN )} of the sequence of N-point DCT blocks {C _k ^(N) } (k = 0,1,… M-1) and the corresponding MN-point DCT There is a transformation matrix A _{(M, N)} of magnitude. This is expressed as in Equation 5 below.

Such analysis in one dimension can be extended in two dimensions. That is, when the DCT block C ^{( LN × MN)} is decomposed into L × M DCT blocks each having a size of N × N, Equation 6 below is used.

The transformation matrix used for decomposition of the DCT block may be given as follows.

The transformation matrix and its inverse matrix are sparse matrices. Therefore, you only need to do fewer multiplications and additions than required for normal matrix multiplication.

라 하면, DCT 블록이 8×8 크기이므로

는 4×4 크기의 네 개의 서브 블록들로 분해된다. ω번째 DCT 블록으로부터 생성된 네 개의 서브 블록은 각각

,

,

,

이며, 간단하게

(p,q=0,1)와 같이 표현할 수 있다.By the same method as described above, the block decomposition unit 120 may decompose the DCT blocks into a plurality of sub blocks, respectively. The compressed image input unit 110 selects the ω th DCT block among the DCT blocks received.

In other words, the DCT block is 8 × 8

Is decomposed into four subblocks of size 4x4. The four subblocks generated from the ωth DCT block are each

,

,

,

Is simply

It can be expressed as (p, q = 0,1).

Four sub-blocks generated from one DCT block are obtained as shown in Equation 7 using Equation 6 defining DCT block decomposition.

Here, matrix A _{(2, 4)} is the same as defined above.

In this case, the size of the subblock may be made smaller so that a larger number of subblocks can be generated from one DCT block. Most preferably.

Next, the brightness information improving unit 130 transforms the value of the transform coefficient corresponding to the DC component among the transform coefficients constituting the sub-block by the first transform function, which is a nonlinear function, in the dark region and the bright region of the original image. Improve brightness information. This process is performed sequentially for a plurality of sub-blocks generated from one DCT block.

Due to the direction or intensity of the light source used when the original image is generated, very bright areas or very dark areas may appear in the original image, and the objects included in these areas lose detail information so that their shape is difficult to recognize. have. The brightness information improving unit 130 is intended to improve the brightness information by restoring the detail information included in the very dark area and the very bright area in the original image. This process is based on the Retinex theory.

Retinex theory separates the illumination and reflectance components of an image and alters these components to increase the visibility of texture detail. Similarly, in the present invention, luminance data of an image is composed of a combination of brightness and detail components. This is because the illuminance component is closely related to the brightness and the reflectance component contributes to the detail of the image. In mathematical terms, each pixel x (i, j) of the image may be expressed as Equation 8 by adding the brightness b (i, j) and the point-to-point of the detail d (i, j).

와 같이 추정된다. 밝기는 영상 전체에 걸쳐 부드럽게 변화한다고 가정하기 때문이다. 반면 DCT 변환 계수의 AC 성분은 영상의 디테일 정보로 표현될 수 있다. 다음으로 디테일 성분의 추정인

는 입력된 휘도 성분인 x(i,j)와 추정된 밝기 성분인

사이의 차에 의해 결정된다.Next, the values of the brightness and detail components are adjusted for the dynamic range and detail of the image. Specifically, the brightness component using the DC component of the DCT conversion coefficient

It is estimated as This is because the brightness changes smoothly throughout the image. On the other hand, the AC component of the DCT transform coefficient may be represented as detail information of an image. Next, an estimate of the detail component

Is an input luminance component x (i, j) and an estimated luminance component

It is determined by the difference between.

는 다음으로 동적 영역 압축, 즉 밝기 정보 개선을 위해 비선형 함수인 G(·)에 의해 변형된다. 그러나 동적 영역 압축만을 기초로 한 영상 복원은 대비가 낮은 장면과 같이 원하지 않는 결과를 초래할 수 있다. 이 러한 문제를 해결하기 위한 것이 영상의 디테일 정보 개선이다. 영상의 대비 개선을 위한 디테일 정보 개선은 추정된 디테일 성분인

에 대해 또 다른 비선형 함수인 F(·)를 적용함으로써 수행된다. 최종적으로 결과 영상 O(i,j)는 변형된 밝기 성분과 개선된 디테일 성분의 합으로 다음의 수학식 9와 같이 얻어진다.The estimated brightness component

Is then transformed by a nonlinear function, G (·), for dynamic domain compression, i.e. brightness information improvement. However, image reconstruction based only on dynamic region compression can have undesirable consequences, such as low contrast scenes. The solution to this problem is to improve the detail information of the image. Detail information enhancement to improve the contrast of the image

This is done by applying another nonlinear function, F (·), to. Finally, the resultant image O (i, j) is obtained as shown in Equation 9 as the sum of the modified brightness component and the improved detail component.

The obtained result image is improved by performing dynamic region compression and detail information on the input image.

As discussed above, image quality improvement in the frequency domain is performed separately for the DC component and the AC component of the image. This process applies equally to the present invention. That is, the present invention improves the quality of the original image by dividing transform coefficients constituting each of the sub-blocks obtained by decomposing the DCT block into DC and AC components, and performing different processes on the respective components.

First, the brightness information improving unit 130 improves brightness information of an original image by applying a nonlinear function such as G (·) to a transform coefficient corresponding to a DC component among transform coefficients constituting each subblock.

The 4 × 4 subblocks generated by the DCT block decomposition by the block decomposition unit 120 include 16 transform coefficients. As described above with reference to FIG. 2, all 64 transform coefficients constituting one DCT block may be classified into 14 frequency bands. When this is applied to a subblock, 16 transform coefficients constituting the subblock are classified into 7 frequency bands according to spatial frequency characteristics.

3 is a diagram illustrating a form of a sub block classified into seven frequency bands. The conversion coefficients are arranged in the order of increasing spatial frequency from the upper left to the lower right as in FIG. 2, and the transform coefficients at positions through which the diagonal lines shown in FIG. 3 pass are included in the same frequency band. The numbers represent the number of frequency bands. The set of seven frequency bands may be represented by Equation 10 below.

Here, φ _n is a set consisting of n frequency bands, and ψ _n is an nth frequency band.

(n=u+v)의 집합으로, 다음 수학식 11과 같이 표현된다. 또한 동일한 주파수 밴드에 포함된 DCT 변환 계수들은 유사한 공간 주파수 특성을 가진다.In addition, each frequency band ψ _n is a transform coefficient constituting the subblock.

A set of (n = u + v), which is represented by the following equation (11). In addition, DCT transform coefficients included in the same frequency band have similar spatial frequency characteristics.

이다.The transform coefficient of the DC component whose value is modified by the brightness information improving unit 130 is one transform coefficient included in a frequency band corresponding to n = 0, that is,

to be.

The improvement of the brightness information of the original image by the brightness information improving unit 130 is for clearly displaying the detailed information of the object included in the dark area and the bright area in the original image. That is, the dynamic range of the original image may be compressed by improving the brightness information.

를 나타내는 것으로 볼 수 있다. 동적 영역을 압축하기 위해 본 발명에서는 영상의 DC 성분의 값을 조절하고 압축된 입력 영상에서 가장 어두운 부분 및 가장 밝은 부분의 세부정보를 개선한다.The purpose of dynamic region compression is to map the natural dynamic region of the signal into smaller regions. This can be done by modifying the local brightness of the image. As mentioned earlier, the brightness of the image changes smoothly throughout the image, so the DC component, the average of each block in the compressed image, is local brightness.

It can be seen as representing. In order to compress the dynamic region, the present invention adjusts the value of the DC component of the image and improves the details of the darkest part and the brightest part of the compressed input image.

과 같이 산출된다. 여기서

은 하나의 DCT 블록으로부터 생성된 서브 블록들 중 pq번째 서브 블록의 DC 성분의 값이다. 다음으로 지역 밝기

는 영상의 어두운 영역에서 비선형 함수를 적용하여

로 변형되는데, 이는 다음의 수학식 14와 같이 표현된다.In the present invention, the average value m of 4 × 4 subblocks, which is a unit used to improve the brightness of the original image, is

It is calculated as here

Is the value of the DC component of the pq-th subblock among subblocks generated from one DCT block. Next, local brightness

Applies a nonlinear function in the dark areas of the image

Is transformed into Equation 14 below.

Here, I _max is the maximum brightness value of the original image (for example, 255 for the 8-bit input image), and γ is an adjustment value determined and input by the user to adjust the degree of improvement of the image. Experimentally, in order to improve the details of the darkest and brightest areas of the image, it is preferable to set γ> 1.

4 is a graph showing a relationship between an input value and an output value when the value of γ is set differently. Referring to FIG. 4, when γ is 1, the input value and the output value are the same, and as the value of γ increases, the low input value is converted into a high output value, and the high input value is converted into a low output value. In other words, the larger the value of γ, the greater the degree of improvement of the brightness component. In this way, the value of gamma may be set to an appropriate value with reference to the degree of improvement of the brightness component.

An image in which brightness information is improved for a dark region of an original image may be expressed as in Equation 13 below.

에 적용되는 양자화 매개변수이다.Here, O ^d (i, j) is an image in which brightness information is improved in a dark region, and Q ^pq (0, 0) is a DC component of the pq-th subblock.

Quantization parameter applied to.

The luminance component of the image is normalized by improving the brightness information in the dark area. Improvement of the brightness information for the bright area is made by the following equation (14), similarly to the dark area.

을 사용하여 밝은 영역을 개선하기 위해, 밝기값을 반전시키고 강조한다. 다음으로, 변형된 값은 최대 밝기값인 I_max로부터 차감되어 밝은 영역으로 다시 변환된 후 압축된다.Here, O ^b (i, j) is an image in which brightness information in a bright area is improved. Equation 14 indicates that the detailed information of the bright area is improved by applying a process similar to that of Equation 13 to the inverse image of the input image. Specifically, given mapping curve

Use to highlight and highlight brightness values to improve bright areas. Next, the modified value is subtracted from the maximum brightness value I _max , converted back to a bright area, and then compressed.

When the brightness information is improved in both the dark and the bright areas, Equations 13 and 14 are combined to obtain an image in which the transform coefficient value corresponding to the DC component is transformed in each subblock as shown in Equation 15 below. Lose.

은 각 서브 블록의 DC 성분에 대해 이상에서 설명한 것과 같은 밝기 정보 개선 과정을 적용하여 얻어진 변형된 DC 성분을 나타낸 것이다.Where Q ⁻¹ [·] represents the inverse quantization process,

Denotes a modified DC component obtained by applying the brightness information improvement process as described above with respect to the DC component of each sub-block.

On the other hand, the value of γ, which is an input value for adjusting the degree of improvement of the brightness information, may be adaptively adjusted according to the content included in the subblock, that is, the amount of high frequency components. By adjusting the value of γ, it is possible to reduce block phenomenon and overshooting phenomenon appearing over a large edge after the image reconstruction process. An enhancement value of the pq-th subblock of the ω-th DCT block among the DCT blocks input to the compressed image input unit 110 is calculated by Equation 16 below.

는 pq번째 서브 블록에 포함된 고주파 성분의 양을 나타내는 표준편차 값이다.

의 값은 서브 블록을 분석하고 큰 에지 주변의 블록 현상을 감소시키기 위해 사용된다.Here, γ is a value input from the user,

Is a standard deviation value representing the amount of high frequency components included in the pq-th subblock.

The value of is used to analyze the subblock and reduce the block phenomenon around the large edge.

의 값은 ω번째 DCT 블록 내의 pq번째 서브 블록에 대한 분산

으로부터 얻어지며,

은 다음의 수학식 17에 의해 산출된다.In equation (16)

Is the variance for the pqth subblock in the ωth DCT block

Obtained from

Is calculated by the following equation (17).

As such, the variance of the subblock may be directly calculated by using transform coefficients corresponding to an AC component among the transform coefficients constituting the subblock.

In addition, the function f (·) included in Equation 16 corresponds to a weighting function for smoothly transitioning from one state to another, and has a trapezoidal response. The weight function f (·) is defined as in Equation 18 below.

Here, τ _x , which are reference values indicating transition to another state, is determined experimentally, and is set to a value capable of processing sub-blocks containing less detail information and noise components included in the sub-block.

을 사용하여 낮은 주파수 밴드에 포함된 강한 에지에 영향을 주지 않고 높은 주파수 밴드 내의 디테일 정보를 적응적으로 개선할 수 있다.The brightness information improving unit 130 is an improvement value individually determined for each sub block by Equation 16 based on the value of γ input from the user.

Can be used to adaptively improve detail information in the high frequency band without affecting the strong edges included in the low frequency band.

는 각각의 서브 블록의 컨텐츠 구성에 의해 개별적으로 결정되는 것으로, 이웃하는 서브 블록들을 고려하지 않으므로 블록 현상을 오히려 심화시킬 우려가 있다. 이를 해결하기 위해

의 값은 입력 영상에서 네 개의 이웃하는 서브 블록들을 기초로 스무딩될 수 있다. 도 5에는 개선값의 스무딩을 위해 사용되는 이웃 서브 블록들의 예가 도시되어 있다.By the way

Since is determined individually by the content configuration of each sub-block, since neighboring sub-blocks are not taken into consideration, there is a concern that the block phenomenon will be deepened. To solve this

The value of may be smoothed based on four neighboring subblocks in the input image. 5 shows an example of neighboring subblocks used for smoothing the improvement value.

라 한다. 그리고 이웃하는 네 개의 서브 블록들을 참조하여 현재 서브 블록의 개선값을 스무딩시켜 얻어진 평균 개선값은 다음의 수학식 19에 의해 얻어진다.The current set of improvement values of the current subblock and its four neighboring subblocks

It is called. The average improvement value obtained by smoothing the improvement value of the current sub block with reference to four neighboring sub blocks is obtained by the following equation (19).

는 ω번째 DCT 블록의 pq번째 서브 블록에 대한 평균 개선값, 1_1×5는 1×5 크기의 모든 원소가 1인 매트릭스, 그리고

는 집합

에 포함된 원소들, 즉 서브 블록들의 개수이다.here,

Is the mean improvement value for the pq-th subblock of the ωth DCT block, 1 _{1 × 5} is a matrix of all elements of size 1 × 5, and 1

Set

Is the number of elements included in the subblock.

As described above, the brightness information improving unit 130 improves the brightness information of the original image by modifying a transform coefficient corresponding to the DC component for each of the plurality of subblocks, thereby detailing the object in the bright and dark areas of the original image. Information can be clear.

The detail information improving unit 140 improves the detail information included in the original image by transforming the value of the transform coefficient corresponding to the AC component among the transform coefficients constituting the sub-block by the second transform function which is a nonlinear function. The detail information improvement may be sequentially performed on a plurality of sub-blocks generated by decomposing the DCT block, and may be performed after the deformation process of the DC component by the brightness information improving unit 130 is performed.

The transform coefficients to be transformed by the detail information improving unit 140 are transform coefficients constituting the remaining frequency bands except for n = 0 frequency bands among the seven frequency bands of the sub-blocks described above. Nonlinear functions such as F (·) apply.

First, the number of transform coefficients included in each frequency band may be expressed as in Equation 20 below.

Here, N (ψ _n ) is the number of transform coefficients included in the n th frequency band.

The spectral content for the nth frequency band in the pqth subblock of the ωth DCT block generated from the original image indicates how many high frequency components are included in the nth frequency band compared to the previous frequency bands. Indicates. The spectral content value for the n th frequency band is defined by the ratio of the high spectral content energy to the low spectral content energy of the frequency bands of the transform coefficients, which is calculated by Equation 21 below.

는 ω번째 DCT 블록의 pq번째 서브 블록 내에서 n번째 주파수 밴드에 대한 스펙트럴 컨텐츠 값이고, ||·||_F ²는 프로베니우스 놈(Frobenius norm)의 제곱이다. 수학식 21의 우변에서 분자는 n번째 주파수 밴드의 평균 에너지를 나타내며, 분모는 n번째 밴드보다 주파수가 낮은 주파수 밴드들의 평균 에너지를 나타낸다.here,

Is the spectral content value for the nth frequency band in the pqth subblock of the ωth DCT block, and || · || _F ² is the square of Frobenius norm. On the right side of Equation 21, the numerator represents the average energy of the n th frequency band, and the denominator represents the average energy of the frequency bands lower than the n th band.

This spectral content value is modified to improve detail information of the original image. In order to change the spectral content value, an enhancement factor is applied to different values according to each subblock.

The enhancement factor λ ^pq for the pq-th subblock is expressed by Equation 22 using the ratio of the transform coefficient value of the transformed DC component to the transform coefficient value of the DC component before transformation in the pq-th subblock. Can be represented.

의 값이 0인 경우를 대비하여 사용되는 값이다.Where δ = 0.01

This value is used in case the value of is 0.

The improvement of detail information of the original image, that is, the improvement of contrast, may be described as a modification by the improvement factor of the spectral content value described above. The modified spectral content value for the nth frequency band in the pqth subblock of the ωth DCT block is defined as in Equation 23 below.

Here, the frequency band where n = 0 represents a DC component, so the range of n is 1 or more.

의 값은 수학식 21에 의해 산출되므로 수학식 21을 수학식 23에 대입하고 정리하면 다음의 수학식 24와 같이 표현된다.Spectral Content

Since the value of is calculated by Equation 21, substituting Equation 21 into Equation 23 and arranging it is expressed as Equation 24 below.

는 다음의 수학식 25와 같이 정의된다.Where the energy ratio of the improved AC component is

Is defined as in Equation 25 below.

의 값은 변형 전의 서브 블록과 AC 성분에 해당하는 변환 계수의 값이 변형된 서브 블록에서 n번째 주파수 밴드보다 낮은 주파수의 주파수 밴드들 전체에 대해 산출된 에너지의 합의 비이다.Energy rain

Is a ratio of the sum of energy calculated for all frequency bands of frequencies lower than the nth frequency band in the subblock before transformation and the subblock in which the value of the transform coefficient corresponding to the AC component is modified.

Finally, the value of the DCT transform coefficient with improved detail information is calculated by the following equation (26).

는 pq번째 서브 블록에서 (u,v)의 위치에 해당하는 변형된 변환 계수의 값,

는 ω번째 DCT 블록의 pq번째 서브 블록에서 n(=u+v)번째 주파수 밴드에 적용되는 개선 인자의 값,

는 ω번째 DCT 블록의 pq번째 서브 블록에서 n(=u+v)번째 주파수 밴드에 대한 에너지 비, 그리고

는 ω번째 DCT 블록으로부터 얻어진 pq번째 서브 블록에서 (u,v)의 위치에 해당하는 변형 전 변환 계수의 값이다.here,

Is the value of the transformed transform coefficient corresponding to the position of (u, v) in the pq-th subblock,

Is the value of the enhancement factor applied to the n (= u + v) th frequency band in the pqth subblock of the ωth DCT block,

Is the energy ratio for the n (= u + v) th frequency band in the pqth subblock of the ωth DCT block, and

Is the value of the transform coefficient before transformation corresponding to the position of (u, v) in the pq-th subblock obtained from the ω-th DCT block.

는 4×4 크기의 서브 블록에 대한 것으로, 8×8 크기의 DCT 블록에 적용되는 경우에 비해 중요도가 크지 않다. 서브 블록의 크기가 작으므로 각 주파수 밴드에 포함된 고주파 성분의 변화가 거의 없다고 볼 수 있기 때문이다. 따라서

의 값을 1로 근사화시키면 수학식 26은 다음 수학식 27과 같이 간소화된다.Energy ratio in (26)

Is for a 4 × 4 subblock, which is not as important as the case where it is applied to an 8 × 8 DCT block. This is because the size of the subblock is small, so that there is little change in the high frequency components included in each frequency band. therefore

When the value of is approximated to 1, Equation 26 is simplified as in Equation 27 below.

That is, if the improvement process of the detail information is simplified by approximating the energy ratio, the transform coefficient value is modified by multiplying the transform coefficient corresponding to the AC component in each sub block by an improvement factor indicating the degree of improvement of the brightness information of the corresponding sub block. can do.

의 값은 서브 블록 내에서 AC 성분에 해당하는 변환 계수의 변형 과정을 종료시키는 기준으로 사용될 수도 있다. 즉, 디테일 정보 개선부(140)는 현재 서브 블록에서 AC 성분에 해당하는 변환 계수의 변형 과정을 수행할 주파수 밴드가 n=4 이상이고

(n=u+v)이면 현재 서브 블록에 대한 디테일 정보 개선 과정을 종료할 수 있다. 고주파 성분에 해당하는 주파수 밴드일수록 디테일 정보의 개선 효과가 크지 않으며, 에너지 비가 이전 주파수 밴드에서 산출된 값과 동일하다면 변환 계수의 값을 변형하여도 개선 전의 영상과 차이가 없기 때문에 계산량을 감소시키기 위해서이다.Meanwhile, energy rain

The value of may be used as a criterion for terminating the transformation process of the transform coefficient corresponding to the AC component in the subblock. That is, the detail information improvement unit 140 has a frequency band for performing a transformation process of a transform coefficient corresponding to an AC component in the current subblock, wherein n = 4 or more.

If (n = u + v), the process of improving detail information on the current subblock may be finished. The higher the frequency band corresponding to the high frequency component, the smaller the improvement effect of detail information. If the energy ratio is the same as the value calculated in the previous frequency band, even if the value of the transform coefficient is modified, there is no difference from the image before the improvement. to be.

The plurality of sub-blocks composed of the DCT transform coefficients modified by the brightness information improvement process and the detail information improvement process described above are reconstructed into one DCT block by applying the DCT block decomposition process described in Equation 7 inversely to improve the image. The process ends.

However, if the value of the transformed transform coefficient is too large compared to the value of the transform coefficient before deformation, side effects such as ringing or block phenomenon may occur around the edge due to excessive improvement of the original image. Therefore, it is desirable to apply a constraint to the value of the transformed transform coefficient so as not to be out of a certain range. Application of the constraint on the transform coefficient is performed by the following equation (28).

는 제약조건이 적용된 변환 계수,

는 변형된 변환 계수이고, 제약조건에 해당하는

및

의 값은here,

Is a transform coefficient with constraints,

Is the transformed transform coefficient, which corresponds to the constraint

And

The value of

와 같이 정의된다.

Is defined as:

The values of α and β are determined experimentally and the most desirable results can be obtained when α = β = 5.

When the values of transform coefficients are transformed for all subblocks generated from DCT blocks, the block reconstruction unit 150 applies transformed DCT by applying an inverse matrix of the transform matrix to subblocks formed of transform coefficients of the transformed value. Reconstruct into blocks.

Subblocks including the transform coefficients to which constraints are applied, such as Equation 28, are reconstructed into 8 × 8 transform DCT blocks, which may be performed by applying Equation 7 inversely. That is, the process of generating one modified DCT block from four sub-blocks is represented by the following equation (29).

Where A _(2,4) is the same matrix as defined above.

For the modified DCT block thus obtained, an additional process may be further performed to eliminate side effects of ringing around a large edge and detail information improvement such as mosquito noise. This is a process of changing the transformed transform coefficient value at the position corresponding to the position where the transform coefficient value is 0 in the input DCT block to 0. That is, the value of the transform coefficient at the same position as the transform coefficient corresponding to 0 in the input DCT block among the transform coefficients constituting the modified DCT block is transformed to 0.

The DCT blocks input to the compressed image input unit 110 are sequentially aligned, and the block decomposition unit 120 decomposes one from the most advanced DCT block to generate subblocks corresponding to the DCT block. At this time, the block decomposing unit 120 decomposes the previous DCT block, and the quality improvement process of the original image is performed by the brightness information improving unit 130 and the detail information improving unit 140, and finally by the block reconstructing unit 150. When a modified DCT block corresponding to the previous DCT block is generated, the DCT block following the previous DCT block is decomposed. That is, after one DCT block is decomposed and a process of improving brightness information and detail information of an original image is sequentially performed on subblocks generated therefrom, subblocks of modified transform coefficients are transformed into modified DCT blocks. When reconstructed, the next DCT block is disassembled and the same process is repeated.

Finally, when a modified DCT block corresponding to all DCT blocks is generated, inverse quantization and inverse DCT are performed on the modified DCT block to obtain a resultant image represented by Equation 30 below. That is, the resultant image is an image in which brightness and detail components are improved in the original image.

6 is a flowchart illustrating a process of performing a preferred embodiment of an image reconstruction method in a compressed region according to the present invention.

Referring to FIG. 6, the compressed image input unit 110 receives a plurality of DCT blocks generated by dividing an original image into a plurality of unit blocks and performing discrete cosine transform (DCT) and quantization transformation on each unit block ( S610). The block decomposition unit 120 decomposes the DCT blocks sequentially arranged one by one by a predetermined transformation matrix to generate a plurality of sub blocks having the same frequency distribution as that of the transform coefficients constituting the DCT block (S620). .

The brightness information improving unit 130 transforms the value of the transform coefficient corresponding to the DC component among the transform coefficients constituting the sub-block by the first transform function, which is a nonlinear function, so that the brightness information in the dark region and the bright region of the original image is changed. To improve (S630). In addition, the detail information improving unit 140 improves the detail information included in the original image by modifying the value of the transform coefficient corresponding to the AC component among the transform coefficients constituting the sub-block by the second transform function which is a nonlinear function ( S640). The modification process of the DC component and the AC component is sequentially repeated for a plurality of sub blocks generated from one DCT block.

When the values of the transform coefficients are transformed with respect to all subblocks generated from the DCT block (S650), the block reconstruction unit 150 applies the inverse matrix of the transform matrix to the subblocks of the transformed transform coefficients, thereby transforming the transformed DCT. Reconstruct into blocks (S660). When the modified DCT block corresponding to all of the input DCT blocks is not generated (S670), the block decomposition unit 120 decomposes the next DCT block to generate sub blocks, and the above-described process is repeated. When a modified DCT block corresponding to all of the plurality of DCT blocks is generated (S670), the quality improvement process of the original image is terminated, and inverse quantization transformation and inverse DCT are performed on the modified DCT blocks, so that brightness information and detail information of the original image are performed. The improved result is obtained.

Experiments were conducted to evaluate the performance of the present invention. The original image for the experiment was a high dynamic range image available on the web. In addition, the initial values of γ and the reference values τ ₁ , τ ₂ , τ _3, and τ ₄ , which are input from the user, are set to 1.95, 0.625, 10, 40, and 80, respectively, for consistency across a plurality of experimental images. Here, as described above, γ is a value input by the user, and the remaining values are empirically obtained to improve small detail information of the image.

의 값이 τ₂와 τ₃ 사이의 범위에 포함됨에 따라 충분히 개선되지만, 평탄한 영역 또는 큰 에지가 존재하는 영역은 상대적으로 덜 개선된다. 나아가 τ₂의 값을 10으로 설정한 것은 주어진 DCT 변환 계수들의 값을 적절하게 조절하기 위한 것이다. 나머지 값들은 비트 연산에 의해 τ₁=(τ₂>>4), τ₃=(τ₂<<2), 그리고 τ₄=(τ₂<<3)과 같이 결정된 것이다. 본 실험에서는 실험 영상이 컬러 영상인 경우에도 휘도 성분만 사용되었다.In other words, dark or light areas in the original image that contain content

While sufficiently improved as the value of falls in the range between τ ₂ and τ ₃ , areas where flat areas or large edges exist are relatively less improved. Furthermore, setting the value of τ ₂ to 10 is to properly adjust the values of given DCT transform coefficients. The remaining values are determined by bit operation as τ ₁ = (τ ₂ >> 4), τ ₃ = (τ ₂ << 2), and τ ₄ = (τ ₂ << 3). In this experiment, only the luminance component was used even when the experimental image was a color image.

In order to evaluate the performance of the present invention, conventional image quality improvement methods such as histogram equalization, α-rooting algorithm, and multi-scale contrast enhancement algorithm are used. Was used for comparison. Among these three techniques, histogram smoothing is a widely used method for improving contrast of an image, and the remaining two algorithms are methods for improving detail information of an image in a compressed region using an AC component similarly to the present invention. The main difference between the present invention, the α-routing algorithm, and the multi-scale contrast enhancement algorithm is that the two components except for the present invention do not modify the DC component value, and the spectral content value of the image defined by Equation 21 is calculated. Is not used.

PSNR and Wang's quality measures were used for the performance comparison between the present invention and the above three techniques. The PSNR numerically represents the difference between the original image and the improved result image, and the King's quality scale represents the characteristics of the image based on human visual perception factors such as block or blurring.

7 is a view showing the results of the image quality improvement according to the conventional method and the present invention. The existing method is a method in which the image enhancement process of the present invention is applied to a DCT block rather than a sub block. In other words, the decomposition process of the DCT block is not used. Looking at the enlarged portion in (a) and (b) of Figure 7, it can be seen that the block phenomenon appeared remarkably along the large edge of the image compared to the case of using the present invention when using the conventional method. This is because the conventional method uses an 8 × 8 DCT block having four times the size of the 4 × 4 subblock used in the present invention. In addition, in the case of using the present invention, detailed information is more clearly expressed than in the case of using the conventional method.

8 is a diagram illustrating an experimental result of applying the present invention to various images. The images positioned on the left side of FIG. 8 are original images, and the resulting images obtained by applying the present invention are shown on the right side of each original image. Referring first to (a) and (b) of FIG. 8, the brightness of the upper left and lower right portions may be improved over the entire image by applying the present invention. In addition, FIG. 8C shows an image in which the right side of the image is brighter than the left side due to sunlight. FIG. 8 (d) obtained by applying the present invention to this image shows a result in which the detail information of the dark region is clearly shown. In particular, the trunk of the tree located on the left side of the image has been improved. 8 (e) shows an image in which the detail information of the dark part is largely lost by the strong light. By applying the present invention, it can be confirmed that the improvement of the structure of the tower can be distinguished as shown in (f) of FIG. Finally, referring to FIGS. 8G and 8H, detailed information of an excessively bright or dark region of an image due to strong sunlight is clearly restored.

In order to measure the quantitative performance of different image quality improvement algorithms, experiments using PSNR and Wang's quality scale were performed on high dynamic range images. The results of applying various algorithms to the same original image are shown in FIG. 9. In the original image shown in (a) of FIG. 9, detail information is lost in a very bright area and a dark area. FIG. 9B is a resultant image obtained by minimum mean brightness error bi-histogram equalization (MMEBHE), which is a form of an improved histogram smoothing algorithm. Areas that are too dark or too bright appear. The image of FIG. 9C obtained by the α-routing algorithm appears almost identical to the original image. In the image of FIG. 9 (d) obtained by a contrast measure algorithm, a block phenomenon is severely shown. The image of FIG. 9E obtained by applying the conventional method using an 8 × 8 DCT block shows better results, but spurious edges occur around a large edge. Finally, referring to FIG. 9 (f), it can be confirmed that, by applying the present invention to the original image, the detail information of the dark region and the bright region is restored more than the previous methods.

Next, MMEBHE (Algorithm 1), α-Routing Algorithm (Algorithm 2), Contrast Measurement Algorithm (Algorithm 3), and 8 x 8 DCT blocks (Algorithm 4) to numerically measure the quality of image enhancement. And the present invention was compared. The performance of the five techniques was measured by PSNR and the King's scale. Each figure is measured using different quantization levels (Q ₃₀ , Q ₅₀ , Q ₇₅ and Q ₉₅ ). The quantization parameter can be adjusted using the quality factor as shown in Equation 31 so that a smaller value can be compressed more.

Tables 1 and 2 below show the performance of each algorithm measured by PSNR and quality measures, respectively. Maximum values from each original image are highlighted in bold.

PSNR (Q ₃₀ / Q ₅₀ / Q ₇₅ / Q ₉₅ ) Algorithm 1 Algorithm 2 Algorithm 3 Algorithm 4 Invention #One 62.6 / 68.4 /
63.0 / 71.8 82.3 / 83.3 /
85.0 / 86.6 68.8 / 68.8 /
68.8 / 68.7 71.3 / 71.3 /
71.4 / 71.4 74.1 / 74.2 /
74.4 / 74.7 #2 71.0 / 71.2 /
72.2 / 73.9 77.1 / 78.0 /
80.1 / 82.9 64.0 / 63.8 /
63.6 / 63.5 67.7 / 67.6 /
67.5 / 67.4 71.8 / 72.1 /
72.6 / 73.5 # 3 68.0 / 68.7 /
68.1 / 68.1 79.4 / 80.4 /
82.4 / 85.4 66.0 / 65.8 /
65.6 / 65.3 67.6 / 67.5 /
67.3 / 67.1 71.2 / 71.3 /
71.6 / 72.4 #4 69.3 / 69.8 /
68.1 / 69.4 74.5 / 74.9 /
75.4 / 77.7 64.0 / 63.6 /
62.9 / 64.1 70.4 / 70.4 /
70.2 / 71.5 72.0 / 72.3 /
73.1 / 73.8 # 5 72.4 / 73.1 /
70.0 / 73.7 79.8 / 80.7 /
82.6 / 85.2 67.3 / 67.1 /
66.9 / 66.7 68.0 / 67.9 /
67.8 / 67.6 70.2 / 70.3 /
70.6 / 71.3 Average 68.7 / 70.2 /
68.3 / 71.4 78.6 / 79.5 /
81.1 / 83.6 66.0 / 65.8 /
65.5 / 65.6 69.0 / 68.9 /
68.8 / 69.0 71.8 / 72.0 /
72.5 / 73.1

WANG (Q ₃₀ / Q ₅₀ / Q ₇₅ / Q ₉₅ ) Algorithm 1 Algorithm 2 Algorithm 3 Algorithm 4 Invention #One 5.8 / 7.3 / 7.5 / 9.1 18.4 / 19.1 /
20.3 / 21.2 17.9 / 18.6 /
19.5 / 20.4 19.3 / 20.1 /
21.2 / 22.3 19.7 / 20.5 /
21.6 / 23.0 #2 7.0 / 7.6 / 8.6 / 9.5 18.2 / 18.8 /
19.7 / 20.6 17.5 / 18.1 /
19.1 / 19.9 18.0 / 18.6 /
19.6 / 20.5 19.1 / 19.7 /
20.8 / 22.0 # 3 5.8 / 6.8 / 8.1 / 9.4 17.9 / 18.7 /
19.9 / 21.0 17.4 / 18.1 /
19.3 / 20.3 17.5 / 18.2 /
19.3 / 20.4 18.5 / 19.3 /
20.7 / 22.3 #4 7.0 / 7.8 / 9.1 / 9.1 18.5 / 19.2 /
20.4 / 20.4 17.9 / 18.6 /
19.9 / 19.7 19.4 / 20.0 /
21.3 / 21.4 19.9 / 20.6 /
22.0 / 21.9 # 5 5.9 / 6.7 / 7.8 / 9.1 17.7 / 18.5 /
19.7 / 20.7 17.2 / 18.0 /
19.1 / 20.0 17.4 / 18.1 /
19.3 / 20.3 17.9 / 18.7 /
20.1 / 21.8 Average 6.3 / 7.2 / 8.2 / 9.2 18.1 / 18.9 /
20.0 / 20.8 17.6 / 18.3 /
19.4 / 20.1 18.3 / 19.0 /
20.1 / 21.0 19.0 / 19.7 /
21.0 / 22.2

The higher the quality of the image, the larger the PSNR and quality scale. Referring to Tables 1 and 2, it can be seen that the α-routing algorithm (algorithm 2) shows the highest value in PSNR, but the present invention shows the highest value in quality measure.

Summarizing the above experimental results, the average of the measured values for each algorithm is shown in Table 3 below.

PSNR WANG quality scale One 2 3 4 Invention One 2 3 4 Invention #One 66.392 84.269 68.696 71.301 74.278 7.384 19.703 19.057 20.693 21.144 #2 72.047 79.489 63.667 67.465 72.441 8.087 19.266 18.598 19.129 20.319 # 3 68.181 81.858 65.603 67.355 71.576 7.484 19.345 18.710 18.832 20.136 #4 69.110 75.585 63.593 70.581 72.733 8.190 19.575 18.989 20.471 21.022 # 5 72.258 82.043 66.961 67.748 70.554 7.316 19.127 18.514 18.739 19.585 Average 69.598 80.649 65.704 68.890 72.316 7.692 19.403 18.774 19.573 20.441

In the case of PSNR, the numerical value for the α-rooting algorithm is the highest, followed by the numerical value for the present invention. However, the PSNR is measured based only on the difference between the original image and the reconstructed result image, and does not measure whether the reconstructed image contains more visual information. On the other hand, the king's quality scale shows the highest value in the present invention. The King's quality scale considers human visual perception factors such as block phenomena and blurring effects when measuring the improvement quality of an image, so that the present invention can produce a satisfactory improvement while bringing relatively small changes to a given image. It means that there is.

The present invention can also be embodied as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and may be implemented in the form of a carrier wave (for example, transmission via the Internet) . The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific preferred embodiments described above, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the claims. Various modifications can be made by those skilled in the art, and such changes are within the scope of the claims.

1 is a block diagram showing the configuration of a preferred embodiment of an image reconstruction apparatus in a compressed region according to the present invention;

2 shows 64 DCT basic functions,

3 is a diagram illustrating a form of a sub block classified into seven frequency bands;

4 is a graph illustrating a relationship between an input value and an output value when the value of γ for adjusting the degree of improvement of an image is set differently;

5 shows an example of neighboring subblocks used for smoothing an improvement value;

6 is a flowchart illustrating a process of performing a preferred embodiment of an image reconstruction method in a compressed region according to the present invention;

7 is a view showing the results of image quality improvement according to the conventional method and the present invention,

8 is a view showing the experimental results of applying the present invention to various images, and

9 is a diagram illustrating a result of applying various algorithms to the same original image.

Claims (19)

A compressed image input unit sequentially receiving a plurality of DCT blocks generated by dividing an original image into a plurality of unit blocks and performing discrete cosine transform and quantization transform on each unit block; A block decomposition unit for decomposing each of the sequentially input DCT blocks by a predetermined transformation matrix to generate a plurality of sub blocks having a frequency distribution equal to a frequency distribution of transform coefficients constituting the DCT block; Brightness information enhancement to improve brightness information in the dark and bright areas of the original image by modifying the value of the transform coefficient corresponding to the DC component among the transform coefficients constituting the sub-block by the first transform function which is a nonlinear function. part; A detail information improving unit for modifying a value of a transform coefficient corresponding to an AC component among the transform coefficients constituting the sub-block by a second transform function which is a nonlinear function to improve detail information included in the original image; And A block reconstruction unit configured to reconstruct a transformed DCT block by applying an inverse matrix of the transform matrix to subblocks formed of transformed transform coefficients when the values of the transform coefficients are transformed with respect to all subblocks generated from the DCT block; Image restoration apparatus comprising a. The method of claim 1, The block resolving unit generates four sub-blocks from the DCT block by Equation A below: Equation A

here,

Is a pq th subblock generated from the ω th DCT block among the plurality of DCT blocks,

Is the ω th DCT block, and A _(2,4) is the transform matrix. The method according to claim 1 or 2, The brightness information improving unit is a modified value of a transform coefficient corresponding to a DC component in the subblock by the first transform function defined by Equation B below.

An image reconstruction device, characterized in that for calculating: Equation B

Here, Q [·] is a quantization conversion function, I _max is the maximum value of the brightness values of the pixels constituting the original image,

Is the transform coefficient corresponding to the DC component of the pq-th subblock generated from the ω-th DCT block,

Is a quantization parameter applied to the quantized DCT transform coefficients, and γ is a preset adjustment value for adjusting the degree of improvement of brightness information. The method of claim 3, wherein And the adjustment value is modified to an improvement value determined by modifying the adjustment value according to the amount of high frequency components included in the sub block so that the adjustment value is uniquely applied to each sub block. The method of claim 3, wherein The brightness information improving unit calculates an improvement value based on the amount of high frequency components included in each sub block for each of the sub blocks generated from the same DCT block, and neighbors the respective sub blocks and the sub blocks. And an average improvement value, which is an average of improvement values determined for a plurality of neighboring sub-blocks, as the adjustment value. The method according to claim 1 or 2, The detail information improvement unit is a modified value of a transform coefficient corresponding to an AC component in the subblock by the second transform function defined by Equation C below.

An image restoring apparatus comprising: calculating Equation C

Is the value of the transform coefficient corresponding to the position of (u, v) in the pq-th subblock generated from the ω-th DCT block,

Is a value of an improvement factor determined based on a degree of deformation of a transform coefficient corresponding to a DC component in a pq-th subblock generated from the ω-th DCT block, and

Is the energy ratio for the n (= u + v) th frequency band in which p + th subblocks generated from the ωth DCT block have the same transform coefficients. The method of claim 6, The value of the improvement factor is determined by the following equation D, the energy ratio is determined by the following equation E: [Equation D]

here,

Is a modified value of the transform coefficient corresponding to the DC component in the pq-th subblock generated from the ω-th DCT block,

Is the value of the transform coefficient corresponding to the DC component of the pq-th subblock generated from the ω-th DCT block, and δ is a value less than or equal to 0.01, [Equation E]

here,

Is a set of transformed values of transform coefficients constituting a second to n-1 th frequency band in a pq th subblock generated from the ω th DCT block,

Is a set of values of transform coefficients constituting a frequency band from the second sub-delta to the n-1 th in the pq-th subblock generated from the ω-th DCT block, and || · || _F ² is the square of the Provenius norm. The method according to claim 1 or 2, If the modified value of the transform coefficients constituting the sub-block is smaller than a preset minimum value, the modified value is replaced with the minimum value. If the modified value is larger than a preset maximum value, the modified value is converted into the maximum value. And image replacement apparatus. The method according to claim 1 or 2, And converting a value of a transform coefficient corresponding to a position of a transform coefficient having a value of 0 in the DCT block into 0 among the transformed transform coefficients constituting the modified DCT block. (a) sequentially receiving a plurality of DCT blocks generated by dividing an original image into a plurality of unit blocks and performing discrete cosine transform and quantization transform on each unit block; (b) generating a plurality of subblocks having a frequency distribution equal to a frequency distribution of transform coefficients constituting the DCT block by decomposing each of the sequentially input DCT blocks by a preset transformation matrix; (c) transforming a value of a transform coefficient corresponding to a DC component among the transform coefficients constituting the sub-block by a first transform function that is a nonlinear function to improve brightness information in a dark region and a bright region of the original image; step; (d) modifying a value of a transform coefficient corresponding to an AC component among transform coefficients constituting the sub-block by a second transform function which is a nonlinear function to improve detail information included in the original image; And (e) reconstructing a transformed DCT block by applying an inverse matrix of the transform matrix to subblocks of transformed transform coefficients when values of the transform coefficients are transformed for all subblocks generated from the DCT block Image restoration method comprising a. The method of claim 10, In the step (b), four sub-blocks are generated from the DCT block by the following equation A: Equation A

here,

Is a pq th subblock generated from the ω th DCT block among the plurality of DCT blocks,

Is the ω th DCT block, and A _(2,4) is the transform matrix. The method according to claim 10 or 11, wherein In the step (c), it is a modified value of a transform coefficient corresponding to a DC component in the subblock by the first transform function defined by Equation B below.

An image restoration method comprising calculating the: Equation B

Here, Q [·] is a quantization conversion function, I _max is the maximum value of the brightness values of the pixels constituting the original image,

Is the transform coefficient corresponding to the DC component of the pq-th subblock generated from the ω-th DCT block,

Is a quantization parameter applied to the quantized DCT transform coefficients, and γ is a preset adjustment value for adjusting the degree of improvement of brightness information. The method of claim 12, And the adjustment value is modified to an improvement value determined by modifying the adjustment value according to the amount of high frequency components included in the sub block so that the adjustment value is uniquely applied to each sub block. The method of claim 12, In step (c), for each of the subblocks generated from the same DCT block, an improvement value is calculated based on the amount of high frequency components included in each subblock, and the respective subblocks and the subblocks are calculated. And an average improvement value, which is an average of improvement values determined for a plurality of neighboring subblocks neighboring to, as the adjustment value. The method according to claim 10 or 11, wherein In the step (d), it is a modified value of a transform coefficient corresponding to an AC component in the subblock by the second transform function defined by Equation C below.

An image restoration method comprising calculating the: Equation C

Is the value of the transform coefficient corresponding to the position of (u, v) in the pq-th subblock generated from the ω-th DCT block,

Is a value of an improvement factor determined based on a degree of deformation of a transform coefficient corresponding to a DC component in a pq-th subblock generated from the ω-th DCT block, and

Is the energy ratio for the n (= u + v) th frequency band in which p + th subblocks generated from the ωth DCT block have the same transform coefficients. The method of claim 15, The value of the improvement factor is determined by the following equation D, the energy ratio is determined by the following equation E: [Equation D]

here,

Is a modified value of the transform coefficient corresponding to the DC component in the pq-th subblock generated from the ω-th DCT block,

Is the value of the transform coefficient corresponding to the DC component of the pq-th subblock generated from the ω-th DCT block, and δ is a value less than or equal to 0.01, [Equation E]

here,

Is a set of transformed values of transform coefficients constituting a second to n-1 th frequency band in a pq th subblock generated from the ω th DCT block,

Is a set of values of transform coefficients constituting a frequency band from the second sub-delta to the n-1 th in the pq-th subblock generated from the ω-th DCT block, and || · || _F ² is the square of the Provenius norm. The method according to claim 10 or 11, wherein If the modified value of the transform coefficients constituting the sub-block is smaller than a preset minimum value, the modified value is replaced with the minimum value. If the modified value is larger than a preset maximum value, the modified value is converted into the maximum value. Image restoration method characterized in that the replacement by a value. The method according to claim 10 or 11, wherein And converting a value of a transform coefficient corresponding to a position of a transform coefficient having a value of 0 in the DCT block into 0 among the transformed transform coefficients constituting the modified DCT block. A computer-readable recording medium having recorded thereon a program for executing the image restoration method according to claim 10 or 11.

Images