KR101868266B1 - Method for deblurring blur image by using sparse representation algorithm and apparatus for the same - Google Patents

Abstract

Disclosed is a deblurring method using a sparse representation algorithm performed in a deblurring apparatus. The method comprises the steps of: obtaining a plurality of prier images based on a sparse representation algorithm from a photographed image including a noise and a blur; estimating a primitive image and a blur kernel variable; and outputting a deblurred image through the estimated primitive image and the blur kernel variable. Accordingly, deblurring can be provided with excellent performance even under an environment having a great image noise.

Description

[0001] The present invention relates to a method and apparatus for blurring image blurring using a sparse representation algorithm,

The present invention relates to a blurred image blurring technique, and more particularly, to a noise blurred image blurred blurring method and apparatus using a sparse rendering algorithm.

Blur of an image can mean image distortion. The causes of image blur are various. Blur is typically a type of image distortion in which the entire image or part of the object is blurred by moving objects within the camera or camera field of view during the exposure time of photographs.

Removing the blur is called image de-blurring. It is assumed that a blur kernel representing a type of blur is known, and a non-blind deblurring that acquires only a blur removed image and a blur kernel and a blur removed image at the same time It can be divided into blind deblurring to be acquired.

In many image information processing fields such as image editing and analysis, the quality of the input image affects the result. Because blur can affect the quality of the input image, detailed study of de-blurring was needed.

Blur occurs frequently in everyday life, and when taking a picture in a dark environment (low light), noise may accompany the blur as well as the image.

However, since conventional debloring algorithms do not consider such noise, they can obtain a satisfactory de-blurred image only when the noise is small.

Specifically, the existing de-blurring algorithms may not satisfactorily degrade the image if only noise of 5% of the input image is added, or even the estimation of the blur kernel may frequently fail.

In order to solve the above problems, an object of the present invention is to provide a method for deblurring images including noise and blur.

According to an aspect of the present invention, there is provided a method of de-blurring using a sparse representation algorithm performed in a de-blurring apparatus, comprising: obtaining an image including noise and blur; Obtaining a plurality of prior image parameters from the photographed image based on a sparse representation algorithm that is a linear combination of a plurality of bases and a rare representation coefficient used for analysis of the photographed image; Obtaining an estimation algorithm including the plurality of fryer image parameters to remove the noise and the blur; Estimating the noise-removed original image from the photographed image using the estimation algorithm; Estimating a blur kernel variable indicating the degree of inclusion of the noise from the estimation algorithm; And outputting a final image when the amount of change of the original image and the blur kernel variable is less than or equal to a predetermined reference value.

Here, the plurality of fryer image variables obtained through the sparse representation algorithm may include a first fryer image variable from which the noise is removed, and a second fryer image parameter from which the noise and the blur are removed.

Here, the first fryer image parameter may be obtained by a sigma sum of the product of the rare representation coefficient and a plurality of first bases including the blur.

Here, the first bases may be obtained through a convolution of the blur kernel variable and a plurality of second bases that do not include the blur required for obtaining the second fryer image variable.

Here, the second fryer image parameter may be obtained by a sigma sum of the product of the rare representation coefficient and a plurality of second bases that do not include the blur.

Here, the rare representation coefficient may be obtained based on a square root error of a plurality of first bases including the blur.

Here, the plurality of bases can be learned from noisy images.

Here, the source image can be obtained by fixing the blur kernel variable in the estimation algorithm.

Here, the blur kernel variable may be obtained by fixing the source image in the estimation algorithm.

According to an aspect of the present invention, there is provided a de-blurring apparatus comprising: a processor; And a memory in which at least one instruction executed via the processor is stored, wherein the at least one instruction is to acquire an imaging image including noise and blur, and to interpret the imaging image A plurality of prior image parameters are obtained from the photographed image based on a rare representation algorithm which is a linear combination of a plurality of bases used and a rare representation coefficient, Estimating an original image from which the noise has been removed from the photographed image using the estimation algorithm, estimating a blur kernel variable indicating the degree of inclusion of the noise from the estimation algorithm, And the amount of change of the original image and the blur kernel variable is determined in advance And outputting the final image if it is below the reference value.

Here, the plurality of fryer image variables obtained through the sparse representation algorithm may be executable to include the noise-removed first fryer image variable and the noise and the second fryer image variable from which the blur is removed.

Here, the first fryer image variable may be executable to obtain a sigma sum of the product of the rare representation coefficient and a plurality of first bases including the blur.

Here, the first bases may be operable to be obtained through convolution of the blur kernel variable with a plurality of second bases that do not include the blur required for obtaining the second fryer image variable.

Here, the second fryer image variable may be executable to obtain a sigma sum of the product of the rare representation coefficient and a plurality of second bases that do not include the blur.

Here, the rare representation coefficient may be executable to be obtained based on a square root error of a plurality of first bases including the blur.

Here, the plurality of bases may be executable to learn from noisy images.

Here, the source image may be executable to be obtained by fixing the blur kernel variable in the estimation algorithm.

Here, the blur kernel variable may be executable to be obtained by fixing the source image in the estimation algorithm.

According to the present invention, a fryer image is generated using a sparse representation algorithm described above in performing image debloring, and the generated fryer image is used in an optimization target expression to generate a noise- There is an effect that the kernel estimation can be performed.

1 is a block diagram showing the structure of a de-blurring apparatus.
FIG. 2 is a flow chart illustrating a debloring method.
3 is a flow chart illustrating a method for de-blurring using a sparse representation algorithm.
4 is a conceptual diagram showing a dictionary (base) including blur.
5 is a conceptual diagram illustrating a method for acquiring a noise-removed first fryer image.
6 is a conceptual diagram showing a method of acquiring a blur and a noise-removed second fryer image.
FIG. 7 is a conceptual diagram showing the relative performance of the de-blurring apparatus using the sparse representation algorithm when the noise is 13%.
8 is a conceptual diagram showing the relative performance of a plurality of de-blurring apparatuses when the noise is 8.5%.
9 is a conceptual diagram showing the relative performance of a plurality of de-blurring apparatuses when the noise is 10%.
10 is a graph showing the maximum signal-to-noise ratio for each noise level of a plurality of de-blurring apparatuses.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the relevant art and are to be interpreted in an ideal or overly formal sense unless explicitly defined in the present application Do not.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In order to facilitate the understanding of the present invention, the same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

In the following, a wireless communication network to which embodiments according to the present invention are applied will be described. The wireless communication network to which the embodiments according to the present invention are applied is not limited to the following description, and the embodiments according to the present invention can be applied to various wireless communication networks.

1 is a block diagram showing the structure of a de-blurring apparatus.

Referring to FIG. 1, the deblurring apparatus 100 may include at least one processor 110 and a memory 120. The deblurring apparatus 100 may further include an input interface device 130, an output interface device 140, a storage device 150, and the like. Each of the components included in the de-blurring apparatus 100 may be connected by a bus 160 to communicate with each other.

The processor 110 may execute program instructions stored in the memory 120 and / or the storage device 150. [ The processor 110 may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which the methods according to the present invention are performed. The memory 120 and the storage device 150 may be composed of a volatile storage medium and / or a non-volatile storage medium. For example, memory 120 may be comprised of read only memory (ROM) and / or random access memory (RAM).

FIG. 2 is a flow chart illustrating a debloring method.

Referring to FIG. 2, the de-blurring apparatus may perform a procedure related to de-blurring by a processor. The de-blurring apparatus can acquire a photographed image including blur caused by noise indicating image noise and shaking of the photographing apparatus (S200).

은 블러가 없는 원시 영상과 블러 커널 변수의 콘볼루션과 잡음

의 합으로 모델링될 수 있다. 해당 수식은 다음과 같은 수학식으로 표현될 수 있다.Specifically, a photographed image including blur and noise

Convolution and Noise of Blur-Free Source Images and Blur Kernel Variables

. ≪ / RTI > The corresponding equation can be expressed by the following equation.

It is common to calculate the regularization term because the image and the blur kernel must be obtained together with a given blur image only. Primarily, we can optimize the form of the following form by adding an organizing term for the raw image and blur kernel variables.

및

는 각각

,

에 해당하는 조직화 항일 수 있다. 원시 영상

은 실제 영상의 잠재 영상(latent image)일 수 있다. 수학식 2의 argmin 함수는

식을 최소화 할 수 있는 변수 값을 도출하는 알고리즘일 수 있다. 디블러링 장치는 원시 영상을 수학식 2의 변수

을 고정 시킨 상태에서

에 대한 식을 풀어서 획득할 수 있다. In Equation 2,

And

Respectively

,

Can be an organizing part corresponding to < RTI ID = 0.0 > Raw video

May be a latent image of the actual image. The argmin function in equation (2)

It can be an algorithm that derives a variable value that minimizes the expression. The de-blurring device transforms the raw image into the variable < RTI ID = 0.0 >

In a fixed state

Can be obtained by solving the equation for.

을 고정 시킨 상태에서

에 대한 식을 풀어서 획득할 수 있다. 디블러링 장치는 원시 영상 및 블러 커널 변수가 미리 설정된 기준 이상으로 변화가 없는 경우 최종 영상을 출력할 수 있다. The de-blurring device uses the blur kernel variable as a variable in equation

In a fixed state

Can be obtained by solving the equation for. The deblurring unit can output the final image if the original image and the blur kernel variable do not change above a preset reference.

을 추정할 수 있다(S210). 디블러링 장치는 추정 알고리즘으로부터 잡음 포함 정도를 지시하는 블러 커널(blur kernel)변수

를 추정할 수 있다(S220).The de-blurring device may obtain an estimation algorithm for removing the blur. The de-blurring device is a noise-

(S210). The de-blurring apparatus includes a blur kernel variable indicating the degree of noise inclusion from the estimation algorithm

(S220).

The deblurring apparatus can determine whether the raw image and blur kernel variable values are equal to or greater than a predetermined reference (S230). The de-blurring apparatus can repeat steps S210-S230 if the amount of change of the original image and the blur kernel variable is less than or equal to a predetermined standard.

The deblurring apparatus can output the final image when the values of the original image and the blur kernel variable are equal to or greater than a predetermined reference (S240).

3 is a flow chart illustrating a method for de-blurring using a sparse representation algorithm.

Referring to FIG. 3, the de-blurring apparatus may generate a rare representation algorithm (or "sparse representation") that includes an organizing term and perform debloring through an estimation algorithm including the generated sparse representation algorithm.

를 기저(basis,

)(또는 사전(dictionary))들의 선형적 결합(linear combination, 결합계수 {

})으로 근사표현을 하는 것을 의미할 수 있다. The sparse representation algorithm uses the given data

(Basis,

(Or dictionary combinations) (linear combination,

}) Can be approximated.

In particular, a convolutional sparse coding (CPC) may represent a full image as a sum of the convolution of the base filter, which is a sparse representation coefficient and an image map.

The coupled coupled concept may mean that there is a mapping between the input source representation and the target representation. The expression for the sparse representation algorithm may be as follows.

Since bases are learned from noise-free images and represent data with a small number of bases, the sparse representation algorithm can be robust to noise itself.

The de-blurring device may perform the de-blurring-related procedure by the processor. The de-blurring device may acquire a photographed image including noise and blur (S300). The plurality of bases can be expressed as follows.

4 is a conceptual diagram showing a base (dictionary) including blur.

에서

는 블러를 포함하지 않은 기저를 지시할 수 있다. 선행 처리 과정에서 자연스러운 촬영 영상을 사용하여 기저가 미리 학습될 수 있다. 블러를 포함하지 않은 기저는 i=1 부터 i=

까지의 값을 가질 수 있다.

는 복수의 기저의 개수를 의미할 수 있다.Referring to FIG. 4, the left side is an image 400 representing a pre-learned base (dictionary).

in

Can point to a base that does not contain blur. The base can be learned in advance by using a natural photographed image in the pre-processing process. The basis without blur is i = 1 to i =

Lt; / RTI >

May mean the number of bases.

에서

는 블러 기저를 지시할 수 있다. 블러 기저는 i=1 부터 i=

까지의 값을 가질 수 있다.And the right side is an image 410 representing a base (hereinafter referred to as "blur basis") (dictionary) including blur.

in

Can indicate a blur basis. The blur basis is i = 1 to i =

Lt; / RTI >

까지의 블러 기저는 다음과 같을 수 있다.A blurred basis can be generated by a convolutional combination of a blur-free basis (hereinafter referred to as a "non-blur basis") and blur. i = 1 to i =

The blur basis to the following may be as follows.

이 블러가 포함된 기저들로 촬영 영상을 희소표현 알고리즘으로 나타낼 수 있다.Blur the bases using the currently obtained blur kernel

The bases containing this blur can represent photographed images with a sparse representation algorithm.

Referring again to FIG. 3, the de-blurring apparatus is configured to extract data from a plurality of fryer images from the photographed image based on a rare representation algorithm approximated by a linear combination of a plurality of bases and a rare representation coefficient, (S310). The noise-removed first fryer image can be obtained as follows.

5 is a conceptual diagram illustrating a method for acquiring a noise-removed first fryer image.

)을 표현한 영상(500)일 수 있다. 오른쪽은 잡음이 제거된 제1 프라이어 영상(

)(510)일 수 있다.Referring to FIG. 5, the left side shows a captured image including noise and blur

(500). On the right side, the first fryer image

) ≪ / RTI >

The formula for obtaining the first fryer image from the photographed image may be as follows.

를 블러 기저(

)와 희소표현 계수(

)의 콘볼루션 값의 시그마 합으로 표현할 수 있다. 희소표현 계수는 다음과 같은 수학식을 통해 획득될 수 있다.The de-

To the blurred basis

) And the rarity coefficient (

) Can be expressed by the sigma sum of convolution values. The scarcity representation coefficient can be obtained by the following equation.

이 최소가 되는

값을 찾는 함수일 수 있다.

는

값을 결정함에 있어서 조절 역할을 하는 파라미터일 수 있다.The argmin function

This is the minimum

It can be a function to find a value.

The

May be a parameter that plays a role in determining the value.

)와 희소표현 계수(

)의 콘볼루션 값의 시그마 합을 잡음이 제거된 제1 프라이어 영상 (

)으로 결정할 수 있다. 블러와 잡음이 제거된 제2 프라이어 영상은 다음과 같이 획득될 수 있다.The de-blurring device has a blurred basis

) And the rarity coefficient (

) Is calculated by subtracting the sigma sum of the convolution values of the first pre-

). The blur and noise-canceled second fryer image can be obtained as follows.

6 is a conceptual diagram showing a method of acquiring a blur and a noise-removed second fryer image.

) 계수 맵을 표현한 영상(600)일 수 있다. 오른쪽은 블러 및 잡음이 제거된 제2 프라이어 영상(

)(610)일 수 있다. 촬영 영상으로부터 제2 프라이어 영상을 얻는 수식은 다음과 같을 수 있다.Referring to FIG. 6, the left side is a rare expression (

(600) representing a coefficient map. On the right is a second fryer image with blur and noise removed (

) ≪ / RTI > The formula for obtaining the second fryer image from the photographed image may be as follows.

)와 희소표현 계수(

)의 콘볼루션 값의 시그마 합을 블러 및 잡음이 제거된 제2 프라이어 영상 (

)으로 결정할 수 있다.The de-blurring device may be a non-

) And the rarity coefficient (

) Of the convolutional value of the second preial image ("

).

Referring again to Figure 3, the de-blurring device may obtain an estimation algorithm that estimates the plurality of fryer images to remove noise and blur. The estimation algorithm can be expressed by the following equation.

과

는

과

로 구체화 시켜 표현한 수학식은 다음과 같을 수 있다.In the experiment,

and

The

and

The following equation can be expressed as follows.

의 데이터 부분,

+

의 조직화 항,

+

, 희소표현 알고리즘으로 구해진 프라이어 영상들(

,

)을 이용한 추가적 조직화 항으로 구성될 수 있다.Each term

Lt; / RTI >

+

, ≪ / RTI &

+

, Fryer images obtained by the sparse expression algorithm (

,

). ≪ / RTI >

)를 추정할 수 있다(S320). 원시 영상 추정은 수학식 10의 변수

를 고정 시킨 상태에서 변수

에 대해서 식을 풀어서

을 획득할 수 있다.The de-blurring device extracts the noise-removed original image (

(S320). The raw image estimation is performed using the variable < RTI ID = 0.0 >

In a fixed state

To solve for

Can be obtained.

)를 추정할 수 있다(S330). 블러 커널 변수 추정은 수학식 10의 변수

를 고정 시킨 상태에서 변수

에 대해서 식을 풀어서

을 획득할 수 있다.The de-blurring device uses a blur kernel variable ("

) (S330). The blur kernel variable estimation is performed using the variable < RTI ID = 0.0 >

In a fixed state

To solve for

Can be obtained.

)를 갱신할 수 있다. 갱신하는 과정은 다음과 같이 표현될 수 있다.The de-blurring device extracts the noise-removed original image (

Can be updated. The process of updating can be expressed as follows.

는 편미분 연산자이고,

는 대응 가중치일 수 있다.Equation (11)

Is a partial differential operator,

May be a corresponding weight.

The deblurring apparatus can determine whether the amount of change of the original image and the blur kernel variable is less than a predetermined standard (S340). If the amount of change of the original image and the blur kernel variable is lower than a predetermined criterion, even if the de-blurring process is repeated, there is almost no difference in the de-blurring performance.

The de-blurring apparatus may repeat steps S310-S340 if the amount of change of the original image and the blur kernel variable is equal to or greater than a predetermined reference value. The deblurring apparatus can output the final image when the amount of change of the original image and the blur kernel variable is equal to or less than a predetermined reference (S350).

An embodiment in which the case of de-blurring using the sparse representation algorithm is compared with the case of not de-blurring may be as follows.

FIG. 7 is a conceptual diagram showing the relative performance of the de-blurring apparatus using the sparse representation algorithm when the noise is 13%.

Referring to FIG. 7, an image 700 having a noise level of 13% is subjected to de-blurring using a sparse representation algorithm and an image 700 having a noise level of 13% An embodiment in which the case of blurring is compared can be confirmed.

The de-blurring device can obtain a sophisticated estimated image and estimated kernel 710 when the captured image 700 having a noise level of 13% is finely de-blurred.

The result measured by the de-blurring method using the sparse representation algorithm may be expressed in the upper right images 720, 722, and 724. The result of the measurement using the de-blurring method without using the sparse representation algorithm may be expressed in the images 730, 732, and 734 at the bottom right.

The image 720 may be a result obtained by performing S310-S340 of FIG. 4 of the present invention once. The image 722 may be a result obtained by performing S310-S340 of FIG. 2 of the present invention twice. The image 724 may be a result obtained by performing S310-S340 of FIG. 2 of the present invention three times.

The image 730 may be a result obtained by performing the S210-S230 process of FIG. 2 of the present invention once. The image 732 may be a result obtained by performing S210-S230 of FIG. 2 of the present invention twice. The image 734 may be a result obtained by performing the steps S210-S230 of FIG. 2 of the present invention three times.

It can be confirmed that noise and blur are more removed as the original image estimation and the blur kernel variable estimation are repeated. Also, when comparing the image 724 and the image 734, it can be confirmed that the performance of the de-blurring method using the rare representation algorithm is excellent.

8 is a conceptual diagram showing the relative performance of a plurality of de-blurring apparatuses when the noise is 8.5%.

Referring to FIG. 8, the upper left corner may refer to an actual image 800 of a villa taken outdoors. And the lower left hand side may mean a shot image 810 including 8.5% noise and blur.

The middle top may be the final image 820, which is a Levin et al. Method, in which a photographed image 810 with 8.5% noise and blur is deblurred. In this case, the peak signal-to-noise ratio (PSNR) may be 7.35 dB.

The maximum signal-to-noise ratio (SNR) is a numerical value of image loss information in lossy compression of an image, and may be a value representing a difference between an original image and a watermarked image when a watermark is inserted into the image.

The maximum signal-to-noise ratio can be calculated using the root mean square error (RMSE) without considering the power of the signal. Since the maximum signal-to-noise ratio is measured on a logarithmic scale, the unit may be in dB.

The maximum signal-to-noise ratio can be high as the loss is small. That is, the larger the maximum signal-to-noise ratio, the more likely it is to output an image similar to an actual image.

And the middle bottom may be a final image 830 obtained by deburring the photographed image 810 including 8.5% noise and blur in the zhong et al scheme. In this case, the maximum signal-to-noise ratio may be 24.55 dB.

The upper right hand side may be a final image 840 which is a perrone and favaro method and which is obtained by deburring a photographed image 810 including 8.5% noise and blur. In this case, the maximum signal-to-noise ratio may be 17.16 dB.

And the lower right end may be the final image 850 obtained by deburring the photographed image 810 including 8.5% noise and blur in the improved de-blurring method of the present invention. In this case, the maximum signal-to-noise ratio may be 26.23 dB.

It can be confirmed that the maximum signal-to-noise ratio of the resulting de-blurring value of the present invention is relatively highest. It can also be confirmed that the result of the debloring of the present invention is relatively similar to the actual image 800. [

9 is a conceptual diagram showing the relative performance of a plurality of de-blurring apparatuses when the noise is 10%.

Referring to FIG. 9, the upper left corner may be an actual image 900 of a face photographed in the open air. And the lower left hand side may mean a shot image 910 including 10% noise and blur.

The middle top may be a final image 920 that has been deburred from the photographed image 910 with 10% noise and blur in a Levin manner. In this case, the maximum signal-to-noise ratio may be 7.00 dB.

And the middle lower end may be a final image 930 obtained by deblurring the photographed image 910 including 10% noise and blur in the vertical mode. In this case, the maximum signal-to-noise ratio may be 26.31 dB.

The upper right hand side may be the final image 940 which is the image 910 with 10% noise and blur in the Peron and Pebbles manner. In this case, the maximum signal-to-noise ratio may be 16.81 dB.

And the lower right end may be a final image 950 obtained by deblurring the photographed image 910 including 10% noise and blur in the improved de-blurring method of the present invention. In this case, the maximum signal-to-noise ratio may be 26.56 dB.

It can be confirmed that the maximum signal-to-noise ratio of the resulting de-blurring value of the present invention is relatively highest. It can also be confirmed that the result of the debloring of the present invention is relatively similar to the actual image 900.

10 is a graph showing the maximum signal-to-noise ratio for each noise level of a plurality of de-blurring apparatuses.

Referring to FIG. 10, the vertical axis indicates the maximum signal-to-noise ratio (PSNR) (dB). And the horizontal axis may indicate a noise level (%). The first line 1000 at the lower left may denote a debloring result value in a naive manner.

And the second line 1010 on the lower left side indicates the result of the debloring without using the fryer image as an embodiment of the present invention of FIG. And the third line 1020 at the lower left may mean a result of debloring using only the second fryer image.

And the fourth line 1030 on the lower left may mean the debloring result using the first and second fryer images of the present invention. And the fifth line 1040 on the lower left can denote the result of the Levening method.

The sixth line 1050 at the bottom left can mean the result of the de-blurring in the Peron and Pebar schemes. The seventh line 1060 at the bottom left can mean the de-blurring result in the Michaeli and irani manner.

The eighth line (1070) at the bottom left can denote the de-blurring result in a longitudinal manner. When comparing the second line 1010, the third line 1020, and the fourth line 1030 in the lower left corner, it can be seen that the maximum signal-to-noise ratio per noise level is higher as the plurality of fryer images are utilized.

The fourth line 1030 on the lower left side indicating the experimental result value of the present invention using the sparse representation algorithm has the best performance in a section where the noise level is 2% or more as compared with the relatively competitive dubbing method .

The methods according to the present invention can be implemented in the form of program instructions that can be executed through various computer means and recorded on 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 computer readable medium may be those specially designed and constructed for the present invention or may be available to those skilled in the computer software.

Examples of computer readable media include hardware devices that are specially 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 the compiler 1, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate with at least one software module to perform the operations of the present invention, and vice versa.

It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined in the appended claims. It will be possible.

Claims (18)

As a deblurring method using a rare representation algorithm performed in a deblurring apparatus,
Obtaining a photographed image including noise and blur;
Obtaining a plurality of prior image parameters from the photographed image based on a sparse representation algorithm that is a linear combination of a plurality of bases and a rare representation coefficient used for analysis of the photographed image;
Obtaining an estimation algorithm including the plurality of fryer image parameters to remove the noise and the blur;
Estimating the noise-removed original image from the photographed image using the estimation algorithm;
Estimating a blur kernel variable indicating the degree of inclusion of the noise from the estimation algorithm; And
And outputting a final image when the amount of change of the original image and the blur kernel variable is equal to or less than a predetermined reference,
Wherein the plurality of fryer image variables obtained through the sparse representation algorithm include a first fryer image parameter from which the noise is removed and a second fryer image parameter from which the noise and the blur are removed. delete The method according to claim 1,
Wherein the first fryer image variable is obtained by a sigma sum of the product of the rare representation coefficient and a plurality of first bases comprising the blur. The method of claim 3,
Wherein the first bases are obtained through a convolution of the blur kernel variable and a plurality of second bases that do not include the blur needed for obtaining the second fryer image variable. The method according to claim 1,
Wherein the second fryer image variable is obtained by a sigma sum of the product of the rare representation coefficient and a plurality of second bases that do not include the blur. The method according to claim 1,
Wherein the rare representation coefficient is obtained based on a square root error of a plurality of first bases comprising the blur. The method according to claim 1,
Wherein the plurality of bases are learned from noise free images. The method according to claim 1,
Wherein the source image is obtained by fixing the blur kernel variable in the estimation algorithm. The method according to claim 1,
Wherein the blur kernel variable is obtained by fixing the source image in the estimation algorithm. As a de-coupling device,
A processor; And
Wherein at least one instruction executed through the processor includes a memory,
Wherein the at least one instruction comprises:
Obtaining a photographed image including noise and blur,
Acquiring a plurality of prior image parameters from the photographed image based on a rare representation algorithm which is a linear combination of a plurality of bases used for analyzing the photographed image and a rare representation coefficient,
Obtaining an estimation algorithm including the plurality of fryer image variables to remove the noise and the blur,
Estimating the noise-removed original image from the photographed image using the estimation algorithm,
Estimating a blur kernel variable indicating the degree of inclusion of the noise from the estimation algorithm, and
And outputting a final image when the amount of change of the original image and the blur kernel variable is equal to or less than a predetermined reference,
Wherein the plurality of fryer image variables obtained through the sparse representation algorithm include a first fryer image variable from which the noise has been removed and a second fryer image parameter from which the noise and the blur have been removed. delete The method of claim 10,
Wherein the first fryer image variable is obtained by a sigma sum of the product of the rare representation coefficient and a plurality of first bases comprising the blur. The method of claim 12,
Wherein the first bases are obtained through convolution of the blur kernel variable with a plurality of second bases that do not include the blur needed for obtaining the second fryer image variable. The method of claim 10,
Wherein the second fryer image variable is obtained by a sigma sum of the product of the rare representation coefficient and a plurality of second bases that do not include the blur. The method of claim 10,
Wherein the sparse representation coefficients are obtained based on a square root error of a plurality of first bases comprising the blur. The method of claim 10,
Wherein the plurality of bases are learned from noise free images. The method of claim 10,
Wherein the source image is obtained by fixing the blur kernel variable in the estimation algorithm. The method of claim 10,
Wherein the blur kernel variable is obtained by fixing the source image in the estimation algorithm.

Images