KR101776501B1 - Apparatus and Method for removing noise using non-local means algorithm - Google Patents

Abstract

An apparatus and method for removing image noise using a non-local means algorithm are disclosed. The apparatus for removing image noise is an apparatus for removing image noise using a local average algorithm which calculates a weight of each pixel in an input image and removes the noise of the input image. The apparatus includes a boundary line detection part for detecting boundary line components of an input image; an activity calculation part for calculating the activity of each pixel of the input image corresponding to the number of the boundary line components; a weight setting part configured to set the weight of each pixel based on the activity of each pixel; and a noise removing part which calculates a restored image where noise for each pixel of the input image are removed by using the weight of each pixel. The weight of each pixel is set in proportion to the activity of each pixel. It is possible to prevent the loss of important information in a boundary region.

Description

[0001] The present invention relates to an apparatus and a method for eliminating image noise using a non-local average algorithm,

Embodiments of the present invention relate to an apparatus and method for eliminating image noise using a local average algorithm that prevents significant information loss in a border region of an image.

Noise in the image occurs during quantization error during the image compression process, in the process of transmitting data, or in the process of acquiring the image through the camera sensor. Since this degrades the performance of the image system, it is essential to remove noise from the image system.

The noise-containing image is expressed by Equation (1) below.

는 원 영상,

는 잡음에 의해 왜곡된 관측 영상,

은 잡음 성분,

는 2차원 영상에서의 화소 위치를 각각 의미한다. 이 때, 일반적으로 잡음 성분

는 0를 평균으로 가지고,

를 분산으로 가진다. here,

In addition,

Is an observation image distorted by noise,

Is a noise component,

Represents a pixel position in a two-dimensional image. At this time, generally,

Has an average of 0,

As a dispersion.

와 잡음 성분

의 정보로부터 복원 영상

를 얻는 과정을 의미한다. 즉, 잡음 제거의 궁극적인 목표는 복원 영상

가 원 영상

과 최대한 동일하도록 산출하는 것이다. 고전적인 잡음 제거 기법으로 평균 필터(averaging filter)를 이용한 잡음 제거 기법, 위너 필터(Wiener filter)를 이용한 잡음 제거 기법, 총 변동(total variation)를 이용한 잡음 제거 기법 등이 있다. 이러한 기법들은 구현하기 쉬고 계산량이 적다는 장점이 있지만 영상의 윤곽선 부분 및 텍스처 부분의 정보들이 손실된다는 치명적인 단점이 있다. Noise cancellation,

And a noise component

From the information of

. ≪ / RTI > That is, the ultimate goal of noise cancellation is reconstruction

The original image

As much as possible. A classical noise cancellation technique is a noise filtering technique using averaging filter, a noise filtering technique using Wiener filter, and a noise filtering technique using total variation. Although these techniques have the advantage of being easy to implement and having a small amount of computation, there is a fatal disadvantage that information of the contour portion and texture portion of the image is lost.

Meanwhile, among the conventional techniques, the noise canceling technique using non-local means has superior performance among various noise canceling methods. Most noise cancellation methods are pixel-by-pixel operations, whereas non-local averaging techniques preserve image details through patch-based operations.

However, the noise canceling technique using non-local means has the same weighting method regardless of the local characteristics of the image, and the local characteristics of the image are not reflected due to the fixed weighting method There is a disadvantage in that an accurate weight value for each pixel can not be guaranteed.

In order to solve the problems of the related art as described above, the present invention proposes an apparatus and method for eliminating image noise using a non-local average algorithm for preventing important information loss in a boundary region of an image.

Other objects of the invention will be apparent to those skilled in the art from the following examples.

According to another aspect of the present invention, there is provided an apparatus for eliminating image noise using a local average algorithm for calculating weights of pixels in an input image and eliminating noise of the input image, A boundary line detecting unit for detecting a boundary line component of the image; An activity calculation unit for calculating an activity of each pixel of the input image corresponding to a somewhat of the boundary line components; A weight setting unit for setting a weight of each of the pixels based on the activity of each of the pixels; And a noise removing unit configured to calculate a restored image by removing noise for each pixel of the input image using the weight of each of the pixels, wherein the weight of each of the pixels is set in proportion to the activity of each of the pixels. Is provided.

The noise calculator may calculate the restored image based on the following equation.

는 상기 입력 영상의 i번째 화소(

)에 대한 상기 복원 화소,

는 상기 i번째 화소(

)를 중심으로 한 검색 윈도우, 상기

는 상기 i번째 화소(

)의 가중치,

는

내에 포함되는 j번째 화소를 각각 의미함. here,

(I-1) th pixel of the input image

The restoration pixel,

(I) th pixel

), A search window centered on

(I) th pixel

),

The

J < / RTI >

The boundary detection unit may calculate a smoothing image by smoothing the input image, and may calculate a boundary line image including the boundary component by applying a Sobel mask to the smoothed image.

The boundary detection unit may calculate the boundary image based on the following equation.

는 상기 경계선 영상,

는 상기 스무딩 영상,

는 가로 방향 소벨 마스크,

는 세로 방향 소벨 마스크,

는 대각선 방향 소벨 마스크,

는 반대 대각선 방향 소벨 마스크,

는 컨볼루션 연산을 각각 의미함.here,

The boundary line image,

The smoothing image,

Lt; RTI ID = 0.0 > Sobel &

A vertical sobel mask,

A diagonal-direction Sobel mask,

Lt; RTI ID = 0.0 > Sobel &

Represents the convolution operation, respectively.

The activity of the pixel can be expressed by the following equation.

는 상기 i번째 화소의 활동성,

는

내에 포함되는 상기 i번째 화소를 중심으로 하는 상기 입력 영상 내지 상기 경계선 영상 내의 사각형의 패치,

는

내의 화소를 각각 의미함. here,

Is the activity of the i-th pixel,

The

A patch of a quadrangle in the input image or the boundary image centered on the i-th pixel included in the patch image,

The

Respectively.

The weight setting unit may set weights of the pixels based on the following equations.

는 상기 j번째 화소(

)를 중심으로 하는 상기 입력 영상 내지 상기 경계선 영상 내의 사각형의 패치,

는 차수 p를 가지는 norm 연산자, 상기

는 상기 입력 영상 내의 잡음의 표준 편차,

및

는 상기 표준 편차와 연관된 기 설정된 파라미터를 각각 의미함.here,

(J) th pixel

A patch of a quadrangle in the input image or the boundary image,

Is a norm operator having degree p,

Is the standard deviation of the noise in the input image,

And

Means a predetermined parameter associated with the standard deviation, respectively.

및

는 아래의 수학식과 같이 표현될 수 있다.remind

And

Can be expressed as the following equation.

According to another embodiment of the present invention, there is provided an image noise reduction method using a local average algorithm which is performed in an apparatus including a processor and calculates weights of pixels in an input image and removes noise of the input image, Detecting a boundary component of the input image; Calculating the activity of each pixel of the input image corresponding to a bit of the boundary component; Setting a weight of each of the pixels based on the activity of each of the pixels; And calculating a restored image in which noise is removed from each pixel of the input image using the weight of each of the pixels, wherein the weight of each of the pixels is set in proportion to the activity of each of the pixels. A noise cancellation method is provided.

According to the present invention, there is an advantage of preventing loss of important information in a boundary region of an image.

FIG. 1 is a diagram for explaining a concept of a noise canceling technique using a conventional non-local means.
2 is a schematic block diagram of an apparatus for removing image noise according to an embodiment of the present invention.
3 is a flowchart illustrating a method of removing an image noise according to an exemplary embodiment of the present invention.
4 is a diagram showing an example of a two-dimensional Gaussian kernel used in the present invention.
5 is a diagram showing an example of a Sobel mask used in the present invention.
6 is a diagram illustrating an input image and a boundary line image according to the present invention.
7 is a view for explaining the concept of activity according to an embodiment of the present invention.
8 is a diagram showing a classification of a norm according to an embodiment of the present invention.

As used herein, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. In this specification, the terms "comprising ", or" comprising "and the like should not be construed as necessarily including the various elements or steps described in the specification, Or may be further comprised of additional components or steps. Also, the terms "part," " module, "and the like described in the specification mean units for processing at least one function or operation, which may be implemented in hardware or software or a combination of hardware and software .

Hereinafter, a conventional noise elimination technique using non-local means will be briefly described, and then a noise elimination apparatus and method according to an embodiment of the present invention will be described.

FIG. 1 is a diagram for explaining a concept of a noise canceling technique using a conventional non-local means.

The noise cancellation technique using non-local means performs noise cancellation as shown in Equation (2) below.

는 관측 영상, 즉 잡음 제거를 수행하기 위한 입력 영상의 i번째 화소(

)에 대한 복원 화소,

는 i번째 화소(

)를 중심으로 한 검색 윈도우(도 1 참조),

는 i번째 화소(

)의 가중치,

는

내에 포함되는 j번째 화소를 각각 의미한다. 그리고,

는 아래의 수학식 3과 같은 조건을 가진다.
here,

Is an i-th pixel of an input image for performing noise removal,

), ≪ / RTI >

Is an i-th pixel (

(Refer to FIG. 1), a search window

Is an i-th pixel (

),

The

Th pixel included in the second pixel. And,

Has the same condition as Equation (3) below.

는 화소

와 화소

의 유사도인

에 의해 산출되며 이는 아래의 수학식 4과 같이 표현된다.
At this time,

The pixel

And pixels

Similarity of

And is expressed by Equation (4) below. &Quot; (4) "

는 가중치를 감소시키는 정도를 결정하는 계수(잡음의 분산

에 의해 결정됨),

는 가중치를 정규화시키는 인자로서, 아래의 수학식 5와 같이 표현된다.
here,

Is a coefficient that determines the degree of weight reduction

Lt; / RTI >

Is a factor that normalizes the weight, and is expressed by Equation (5) below.

는 아래의 수학식 6과 같이 표현된다.
At this time, the concept of a patch is introduced to calculate the similarity between two pixels. A patch is a set of pixels in the form of a square centering on an arbitrary pixel (see FIG. 1). a < a > -type patch generated for the i-th pixel

Is expressed by Equation (6) below.

와 화소 와의 유사도는 화소

를 중심으로 하는 패치

와, 화소

를 중심으로 하는 패치

와의 유사도에 의존한다. 그리고 유사도를 측정하는 방법으로 아래의 수학식 7과 같이 L₂ norm 거리(distance)을 사용한다.
In the non-local average method,

And pixels ≪ / RTI >

Patch centered on

And pixels

Patch centered on

. As a method of measuring the similarity, L ₂ norm distance is used as shown in Equation (7) below.

는 항상 음수 값을 가지며, 따라서,

와

사이의 L₂ norm 거리가 작을수록 유사도

가 커진다. 더불어, 패치간의 유사도가 클수록 수학식 4에 의해 화소

에 부여되는 가중치로 커지게 된다. here,

Lt; / RTI > always has a negative value,

Wow

The smaller the L ₂ norm distance between

. Further, as the degree of similarity between patches is larger,

As shown in FIG.

와 패치

사이의 유사도가 매우 크기 때문에 화소

에서는 높은 가중치를 가지게 된다. 반면에 패치

와 패치

사이의 유사도는 적기 때문에

에 부여된 가중치는 더 낮은 값을 가진다.
Referring to Figure 1,

And patch

The degree of similarity between the pixels

We have high weights. On the other hand,

And patch

The degree of similarity between

Lt; / RTI > has a lower value.

2 is a schematic block diagram of an apparatus for removing image noise according to an embodiment of the present invention.

2, the apparatus 200 for eliminating image noise according to an exemplary embodiment of the present invention includes an edge detection unit 210, an activity calculation unit 220, A weight setting unit 230, and a noise removing unit 240.

3 is a flowchart illustrating a method of removing an image noise according to an exemplary embodiment of the present invention. Here, the method may be performed in an apparatus including a processor.

Hereinafter, the function of each component and the process performed for each step will be described in detail with reference to FIG. 2 and FIG.

First, in step 310, the boundary detection unit 210 detects an edge component of an input image (observation image).

According to one embodiment of the present invention, the boundary detection unit 210 calculates a smoothing image by smoothing the input image, and applies a sobel mask to the smoothing image to generate a boundary image including the boundary component Can be calculated.

More specifically, the boundary detection unit 210 may calculate a smoothing image using Equation (7) below.

는 스무딩 영상,

는 가우시안 커널,

는 입력 영상,

는 컨볼루션 연산을 각각 의미한다. 한편, 도 4에서는 2차원의 가우시안 커널의 일례를 도시하고 있으며, 이는 아래의 수학식 9과 같이 표현될 수 있다.
here,

A smoothing image,

Gaussian kernel,

The input image,

Represents a convolution operation, respectively. FIG. 4 shows an example of a two-dimensional Gaussian kernel, which can be expressed by Equation (9) below.

Here, the ordered pair (x, y) represents the horizontal and vertical positions with the origin of the mask center.

The boundary detection unit 210 may calculate the boundary image according to Equation (10) using the boundary image calculated as described above.

는 경계선 영상,

는 가로 방향 소벨 마스크,

는 세로 방향 소벨 마스크,

는 대각선 방향 소벨 마스크,

는 반대 대각선 방향 소벨 마스크를 각각 의미한다. here,

However,

Lt; RTI ID = 0.0 > Sobel &

A vertical sobel mask,

A diagonal-direction Sobel mask,

≪ / RTI > refers to the opposite diagonal direction Sobel mask, respectively.

The Sobel mask is the same as that shown in Fig. 5, and the input image (Fig. 6 (a)) and the boundary line image (Fig. 6 (b)) calculated therefrom are shown in Fig.

Next, in step 320, the activity calculating unit 220 calculates the activity of each pixel of the input image corresponding to a somewhat of the boundary line components. At this time, the activity of the pixel can correspond to the activity of the square patch centered on the pixel.

That is, the greater the information existing in the patch, the higher the activity. In the present invention, it is defined that the activity is high when the image has a large border line component, and the less active the border line component is.

FIG. 7 is a view for explaining the concept of activity according to an embodiment of the present invention. It can be seen that the patches having many borderline components have greater activity than the patches having few borderline components.

The activity of the pixel, that is, the activity of the patch centered on the pixel, can be expressed by Equation (11) below.

는 i번째 화소(i번째 화소를 중심으로 하는 패치)의 활동성,

는 경계 영역

내에 포함되는 i번째 화소를 중심으로 하는 정사각형의 패치,

는

내의 화소를 각각 의미한다. here,

Is the activity of the i-th pixel (the patch centering on the i-th pixel)

Lt; RTI ID =

A square patch centered on the i-th pixel included in the pixel,

The

Respectively.

Subsequently, in step 330, the weight setting unit 230 sets the weight of each of the pixels based on the activity of each of the pixels (patches).

That is, in order to give accurate weights reflecting the local characteristics of the image, the weight values described above are modified in the present invention.

In more detail, the weights should satisfy two characteristics. That is, in the case of i) active pixels (patches), a relatively higher weight should be ensured in order to preserve boundary components (Constitution 1), ii) strong noise in case of high noise variance value, (Constitution 2).

Accordingly, the weight setting unit 230 according to an embodiment of the present invention can set the weight of each pixel (patch) in proportion to the activity of each pixel. This can be expressed as Equation (12) below.

는 검색 영역

내에 포함되는 j번째 화소

를 중심으로 하는 정사각형의 패치,

는 차수 p를 가지는 norm(norm) 연산자,

는 입력 영상 내의 잡음의 표준 편차,

및

는 표준 편차와 연관된 기 설정된 파라미터를 각각 의미한다. here,

The search area

J < th > pixel

A square patch centered on the center,

(Norm) operator with degree p,

Is the standard deviation of the noise in the input image,

And

Quot; means a predetermined parameter associated with the standard deviation, respectively.

및

는 아래의 수학식 13과 같이 표현될 수 있다.
in this case,

And

Can be expressed by Equation (13) below.

및

를 잡음의 분산 값에 비례하도록 설계한다. Referring to the above equation, the norm (L ₄ norm) with degree p of 4 is used for the high activity region in the input image, and the norm (L ₁ norm) with the degree p of 1 is used in the low activity region Thereby satisfying the characteristic of the constitution 1. Then, in order to satisfy the characteristic of Configuration 2

And

Is designed to be proportional to the variance of the noise.

8 is a diagram showing a classification of a norm according to an embodiment of the present invention. In FIG. 8, the white region is the L ₄ norm, the gray region is the norm (L ₂ norm) where the degree p is 2, and the black region is the portion to which the L ₁ norm is applied.

Lastly, in step 340, the noise removing unit 240 may calculate a restored image from which noise has been removed for each pixel of the input image using the weight of each pixel. At this time, Equation (2) can be applied.

In summary, the noise removal method using non-local averaging depends on the accuracy of the weight given to each pixel (patch). Since the conventional method uses a consistent norm as the L ₂ norm, The characteristics are not reflected and it is difficult to guarantee the accuracy of the weights. Therefore, the noise canceling apparatus 200 and method using non-local averaging according to an exemplary embodiment of the present invention can calculate three weights of L ₁ , L ₂ , and L ₄ , which are not consistent norms for calculating weights reflecting the local characteristics of the image norm is used to prevent important information loss in the boundary region of the image.

In addition, the above-described technical features may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and limited embodiments and drawings. However, it should be understood that the present invention is not limited to the above- Various modifications and variations may be made thereto by those skilled in the art to which the present invention pertains. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

Claims (9)

An image noise removal apparatus using a non-local average algorithm by calculating weights of pixels in an input image,
A boundary line detector for detecting a boundary component of an input image;
An activity calculation unit for calculating an activity of each pixel of the input image corresponding to a somewhat of the boundary line components;
A weight setting unit configured to set a weight of each of the pixels in proportion to the activity of each of the pixels; And
And a noise removing unit for calculating a restored image from which noises are removed for each pixel of the input image using the weights of the pixels,
Wherein the weight setting unit sets a weight of each of the pixels based on the following equation.

here,

The weight of the i-th pixel of the input image,

Is a search window centered on the i-th pixel,

The

A patch of a quadrangle in the boundary image including the input image or the boundary component centered on the i-th pixel included in the patch image,

The

A patch of a quadrangle in the input image or the boundary image centered on the jth pixel included in the boundary image,

Is the norm operator with degree p,

Is the standard deviation of the noise in the input image,

Is the activity of the i-th pixel,

And

Means a predetermined parameter associated with the standard deviation, respectively. The method according to claim 1,
Wherein the noise eliminator calculates the reconstructed image based on the following equation.

here,

Is the restored image for the i-th pixel,

Denotes the j-th pixel. 3. The method of claim 2,
Wherein the boundary detection unit calculates a smoothing image by smoothing the input image, and calculates the boundary image by applying a Sobel mask to the smoothed image. The method of claim 3,
Wherein the boundary detection unit calculates the boundary image based on the following equation.

here,

The boundary line image,

The smoothing image,

Lt; RTI ID = 0.0 > Sobel &

A vertical sobel mask,

A diagonal-direction Sobel mask,

Lt; RTI ID = 0.0 > Sobel &

Represents the convolution operation, respectively. The method of claim 3,
Wherein the activity of the pixel is expressed by the following equation.

here,

The

Respectively. delete The method according to claim 1,
remind

And

Is expressed by the following equation. ≪ EMI ID = 1.0 >

A method for removing image noise using a non-local average algorithm, the method being performed in a device including a processor,
Detecting a boundary component of an input image;
Calculating the activity of each pixel of the input image corresponding to a bit of the boundary component;
Setting a weight of each of the pixels in proportion to the activity of each of the pixels; And
And calculating a restored image from which noise has been removed for each pixel of the input image using the weights of the pixels,
Wherein the step of setting the weights sets the weights of the pixels based on the following equation.

here,

The weight of the i-th pixel of the input image,

Is a search window centered on the i-th pixel,

The

A patch of a quadrangle in the boundary image including the input image or the boundary component centered on the i-th pixel included in the patch image,

The

A patch of a quadrangle in the input image or the boundary image centered on the jth pixel included in the boundary image,

Is the norm operator with degree p,

Is the standard deviation of the noise in the input image,

Is the activity of the i-th pixel,

And

Means a predetermined parameter associated with the standard deviation, respectively. 9. The method of claim 8,
Wherein the step of removing noise comprises calculating the restored image based on the following equation.

here,

(I-1) th pixel of the input image

The restored image,

Denotes the j-th pixel.

Images