KR100996036B1 - Method and appratus for deblurring in image - Google Patents

Abstract

In the blur removal method of an image, a process of estimating a blur of an input image, a process of modeling a blur estimation error generated during the blur estimation as noise, and a minimization function having a normalization function as a normalization parameter Estimating a normalization function to reflect an error value of the blur of each channel, and performing an iteration process until it satisfies a preset termination condition, according to the error value of the blur of each channel. Updates a function, including removing blur using a minimize function.

Blur, Image restoration, Noise

Description

Blur removing method and device for image {METHOD AND APPRATUS FOR DEBLURRING IN IMAGE}

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image processing, and more particularly, to a method and apparatus for removing blur generated in an image acquisition process in order to improve image quality.

Blur of the image is caused by a camera's mismatch of focus, long-term CCD exposure, and camera shake in an interlaced scan camera. In addition, even in the case of CCD exposure for a short time, when an object near the camera moves very quickly, blur may occur as in the case of a long time CCD exposure.

Blur has a direct effect on the clarity that has the greatest impact on the quality of the image. These blurs are not easily removed and cause various artifacts. Recently, a method of increasing the resolution while eliminating the blur from several images and recognizing the limitation of removing the blur from one image has been studied. However, even in this case, if the blur is not accurately estimated, the blur affects like noise, and thus the deterioration of the final image cannot be avoided.

Image techniques to remove blur and noise have been studied under the name of image restoration. An image model for explaining this may be represented by Equation 1 below.

In Equation 1, x is an original image to be restored, y is an acquired image, H is blur, and n is noise.

A series of processes for obtaining the original image x to be restored from the acquired image y is called image restoration. In particular, obtaining a single high resolution image from several images is called superresolution / high-resolution image reconstruction. This can be expressed as Equation 2 below.

In Equation 2, x is an original image to be reconstructed, y _{1 to k} are acquired images, H _{1 to k} are blurs of each image, and n is noise.

Referring to Equation 2, the blur is different according to each of various sensors. Each blur is represented by H ₁ to H _k . A method of synthesizing this is shown in Equation 3 below.

The solution of Equation 3 can be obtained as Equation 4 below.

The above-developed schemes assume that the estimation of the blur is accurate. In practice, however, it is almost impossible to accurately estimate blur. Therefore, some errors of blur estimation must be taken. Therefore, a more accurate blur removal method considering the error of the blur estimation is required.

An object of the present invention is to provide a method and apparatus for removing blur from a plurality of motionless images in order to improve image quality.

According to an aspect of the present invention, there is provided a blur removal method of an image, comprising: estimating a blur of an input image, modeling a blur estimation error generated at the blur estimation as noise, and Estimating a minimization function having a normalization function as a normalization parameter, estimating the normalization function to reflect an error value of the blur of each channel, and iterating until a predetermined termination condition is satisfied And updating the normalization function according to the error value of the blur of each channel through the process, and removing the blur by using the minimization function.

According to another aspect of the present invention, in a blur removal apparatus of an image, a blur estimator for estimating a blur of an input image and modeling a blur estimation error generated during the blur estimation as noise, A normalization function estimator for estimating a normalization function reflecting the error value of the blur, and updating the normalization function according to the error value of the blur of each channel through an iteration process until a preset termination condition is satisfied, And a blur removal unit for removing blur using a minimization function having a normalization function as a normalization parameter, and an image reconstruction unit for outputting a result reflecting the blur removal.

The present invention has an effect of obtaining a high resolution image by efficiently eliminating blur by receiving a plurality of images as inputs for blur removal, estimating blur and reflecting an error generated in the blur estimation process in the blur removal process.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an apparatus and an operation method of the present invention will be described in detail with reference to the accompanying drawings. In the following description, specific matters such as specific elements are shown, which are provided to help a more general understanding of the present invention. It is self-evident to those of ordinary knowledge in Esau. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

The present invention performs an operation for removing blur from several motionless images and proposes a method and apparatus for efficiently removing blur by automatically considering an error of the estimated blur in this process.

1 is a flowchart illustrating a blur removal operation of an image according to an exemplary embodiment.

First, when the image is input, the blur of the input images is estimated in step 110 and the blur estimation error is modeled as noise. Next, in step 120, the normalization function reflecting the blur estimation error is estimated, and in step 130, the minimization function having the normalization function as a normalization parameter is estimated to remove the blur. In step 140, it is determined whether the termination condition is satisfied. If the end condition is satisfied as the result of the determination, the process terminates. If the end condition is not satisfied, the process proceeds to step 120 and iterates through the normalization function to continuously upgrade. In this process, the normalization function automatically changes according to the blur estimation error and outputs an optimal result. The blur removal operation will be described in more detail.

<Model of Noise>

The estimation error of the blur can be modeled as noise. If the correct estimated blur is called H _true and the incorrectly estimated blur is called H _false , the image model of Equation 1 may be changed to the same as the image model of Equation 5 using the estimated H.

In Equation 5, n is noise of an image, y is an acquired image, x is an original image, H _true is an estimated blur accurately, H _false is an error of the estimated blur, and n is noise.

According to Equation 5, the total noise may be modeled as in Equation 6 below.

　

　

In Equation 6, n _T is the total noise of the image, y is the acquired image, x is the original image, H _true is the estimated blur accurately, H _false is the error of the estimated blur, and n is the noise. .

As a result, the variance of the total noise n _T may be defined as in Equation 7 below.

In Equation 7, n _T is the total noise of the image, x is the original image, H _false is the estimated blur error, and n is the noise of the image.

Thus, the overall noise is leveled by the estimation error H _false of the blur.

<Minimization functional for blur removal>

Various methods can be used to remove the blur. In an embodiment of the present invention, a Constraint least square method is used.

는 정규화 파라미터이고, C는 고역필터이다.In Equation 8, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization parameter and C is the high pass filter.

과 블러를 제거하는 과정에서 발생하는 링잉을 줄이는 부분

의　사이를 조절하는 값이다. 그러나 각 채널마다 블러 추정 에러가 다르기 때문에, 본 발명은 정규화 파라미터의 위치를 하기의 수학식 9와 같이 수정한다.In Equation 8, the normalization parameter α is a part for removing blur

Reduces ringing incurred while removing excess and blur

This value is to adjust between. However, since the blur estimation error is different for each channel, the present invention modifies the position of the normalization parameter as shown in Equation 9 below.

는 정규화 파라미터이고, C는 고역필터이다.In Equation 9, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization parameter and C is the high pass filter.

를 채널에 특성에 따라 변하는 정규화 함수

로 수정한다. 따라서 전체적인 최소화 함수(Minimization functional)는 하기의 수학식 10과 같이 수정된다. In Equation 9, a normalization parameter for acquiring an area according to the blur estimation error of each channel.

Normalization function that varies with the characteristics of the channel

Correct with Therefore, the overall minimization function (Minimization functional) is modified as shown in Equation 10 below.

는 정규화 함수이고, C는 고역필터이다.In Equation 10, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function and C is the high pass filter.

<Definition of Normalization Function>

The normalization function according to an embodiment of the present invention must satisfy three characteristics as follows.

First, the normalization function must be changed by the value of the error of the blur of each channel, and the normalization function must increase as the noise modeled by the blur estimation error of each channel increases. Second, in the process of minimizing Equation 10, the normalization function should be differentiable. If no differentiation is possible, there is always no guarantee of the solution. Third, the normalization function should change automatically according to the blur estimation error.

In order to satisfy the above three characteristics, in the present invention, a normalization function is defined as in Equation 11 below.

는 정규화 함수이고, P는 상수값이다.In Equation 11, y is an acquired image, x is an original image, H is a blur of an image,

Is a normalization function and P is a constant value.

<Remove Blur>

After the blur estimation for each channel, the normalization function for each channel is defined as shown in Equation 11 and substituted into Equation 10 to differentiate by x to obtain a solution.

는 정규화 함수이고, C는 고역필터이고, C^T는 C의 전치행렬이다.In Equation 12, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function, C is the high pass filter, and C ^T is the transpose of C.

According to Equation 12, an iteration equation may be written as Equation 13 below.

는 상기 수학식 12의 값이다.In Equation 13, n is an iteration number, β is an iteration step, and

Is the value of Equation 12 above.

<Termination Condition>

In the blur removal method according to an embodiment of the present invention, a stop condition is set as in Equation 14 below.

2 is a block diagram of an apparatus for removing blur of an image according to an exemplary embodiment. Referring to FIG. 2, an apparatus for removing blur of an image according to an exemplary embodiment may include a blur estimator 210, a normalization function estimator 220, a blur remover 230, and an image reconstructor 240. Include.

The blur estimator 210 estimates the blur and models the blur estimation error as noise.

The normalization function estimator 220 estimates a normalization function reflecting the blur estimation error of each channel.

The blur remover 230 updates the normalization function according to the blur estimation error of each channel and removes the blur through an iteration process using a minimization function having the normalization function as a normalization parameter.

The image reconstructor 240 outputs a result reflecting the blur removal.

As described above, an operation and a configuration of an image blur removing method and apparatus according to an embodiment of the present invention can be made. Meanwhile, in the above description of the present invention, a specific embodiment has been described. It can be carried out without departing.

1 is a flowchart illustrating an operation of removing blur of an image according to an exemplary embodiment.

2 is a block diagram of an image blur removal apparatus according to an embodiment of the present invention.

Claims (16)

In the blur removal method of the image, Estimating the blur of the input image; Modeling a blur estimation error occurring in the blur estimation as noise; Estimating a minimization function having the normalization function as a normalization parameter, Estimating the normalization function to reflect an error value of the blur of each channel; And updating the normalization function according to the error value of the blur of each channel through an iteration process until a predetermined termination condition is satisfied, and removing the blur by using the minimization function. How to remove blur in the image. The method of claim 1, wherein the noise modeled by the blur estimation error is represented by Equation 6 below, and the variance of the noise is represented by Equation 7 below. &Quot; (6) &quot;

　 [Equation 7]

In Equation 6, 7, n _T is the total noise of the image, x is the original image, y is the acquired image, H _true is the estimated blur accurately, and H _false is the estimated blur. Error, and n is noise. 3. The method of claim 2, wherein the minimization function is defined as in Equation 10 using a Constraint least square method. [Equation 10]

In Equation 10, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function and C is the high pass filter. The method of claim 3, wherein the normalization function, Varies with the blur estimation error of each channel and accordingly with the noise modeled by the blur estimation error of each channel, Differential is possible in the process of minimizing the minimization function, And blur the image automatically according to the blur estimation error. 4. The method of claim 3, wherein the normalization function is defined as in Equation 11 below. [Equation 11]

In Equation 11, y is an acquired image, x is an original image, H is a blur of an image,

Is a normalization function and P is a constant value. 6. The method of claim 5, wherein substituting the normalization function into the minimization function to differentiate by x is as follows. [Equation 12]

In Equation 12, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function, C is the high pass filter, and C ^T is the transpose of C. The method of claim 6, wherein the iteration equation is represented by Equation 13 below. [Equation 13]

In Equation 13, n is an iteration number, β is an iteration step, and

Is the value of Equation 12 above. delete In the blur removal apparatus of the image, A blur estimator for estimating a blur of an input image and modeling a blur estimation error occurring in the blur estimation as noise; A normalization function estimator for estimating a normalization function reflecting an error value of the blur of each channel; The normalization function is updated according to an error value of the blur of each channel through an iteration process until an end condition is satisfied, and a blur agent is removed by using a minimization function having the normalization function as a normalization parameter. With rejection, And an image restoring unit outputting a result of reflecting the blur removal. The apparatus of claim 9, wherein the noise modeled by the blur estimation error is represented by Equation 6 below, and the variance of the noise is represented by Equation 7 below. &Quot; (6) &quot;

　 [Equation 7]

In Equation 6, 7, n _T is the total noise of the image, x is the original image, y is the acquired image, H _true is the estimated blur accurately, and H _false is the estimated blur. Error, and n is noise. 12. The apparatus of claim 10, wherein the minimization function is defined as in Equation 10 using a constraint least square method. [Equation 10]

In Equation 10, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function and C is the high pass filter. The method of claim 11, wherein the normalization function, Varies with the blur estimation error of each channel and accordingly with the noise modeled by the blur estimation error of each channel, Differential is possible in the process of minimizing the minimization function, And the image is automatically changed according to the blur estimation error. The apparatus of claim 11, wherein the normalization function is defined as in Equation 11 below. [Equation 11]

In Equation 11, y is an acquired image, x is an original image, H is a blur of an image,

Is a normalization function and P is a constant value. 15. The apparatus of claim 13, wherein substituting the normalization function into the minimization function to differentiate by x is as shown in Equation 12 below. [Equation 12]

In Equation 12, y is an acquired image, x is an original image, H is a blur of an image,

Is the normalization function, C is the high pass filter, and C ^T is the transpose of C. The apparatus of claim 14, wherein the iteration equation is represented by Equation 13 below. [Equation 13]

In Equation 13, n is an iteration number, β is an iteration step, and

Is the value of Equation 12 above. delete

Images