KR20080044737A - Method for processing an ultrasound image - Google Patents

Abstract

A method for processing an ultrasonic image is provided to reinforce an edge by using an edge application filter based on a wavelet transform and a normalize edge map, reduce a speckle noise, and restrain a flicker phenomenon in an edge peripheral part. The 1 level wavelet decomposition of a ultrasonic image is performed(S110). A weighted edge map is extracted by using eigen-values of a tangent direction and a normal direction in respective pixels of the wavelet-decomposed image(S120). An independent edge is removed from the weighted edge map(S140). Adaptive filtering is performed by applying inter-different weight values according to pixel values of respective edges in the weighted edge map(S150). The wavelet synthesis of the adaptively filtered wavelet decomposition image is performed(S160).

Description

Method for processing an ultrasound image

1 is a flowchart illustrating a process of improving image quality of an ultrasound image according to the present invention.

Figure 2 is a photograph showing an example showing the weighted edge map calculated in the ultrasound image in three dimensions in accordance with the present invention.

3 is a photograph showing an example in three dimensions of a weighted edge map obtained by applying a median filter to the weighted edge map obtained in FIG.

4 is an exemplary diagram illustrating a linear interpolation method of adjacent pixel values using eigenvectors and directionality at edges.

5 is a graph showing the conversion of weights by a gamma function and a linear increase function.

6 is a graph showing the frequency response of a filter used in the tangential direction in directional filtering.

7 is a glyph showing the frequency response of a filter used in the normal direction in directional filtering.

8 is a photograph showing a circular ultrasound image of the liver.

FIG. 9 is a photograph showing an example of an ultrasound image in which filtering according to the present invention is applied to a circular ultrasound image of the liver of FIG. 8; FIG.

10 is a photograph showing the original ultrasound image of the fetus.

11 is a photograph showing an example of an ultrasound image in which filtering according to the present invention is applied to the original ultrasound image of the fetus of FIG. 10.

The present invention relates to an image processing method, and more particularly, to an ultrasonic image processing method for improving the quality of an ultrasonic video by using a wavelet transform and a weighted edge map.

Recently, a diagnostic apparatus using ultrasound has been widely used in the general medical field. The ultrasound imaging apparatus is an apparatus that transmits ultrasound to an object and then detects a reflected wave returning from the object and provides an image obtained therefrom. However, the ultrasonic waves returned from the object are simultaneously reflected and scattered by the medium and the small living tissue. The noise generated due to this is called speckle noise. This speckle noise acts as a major cause of deterioration of the ultrasound image quality, which makes accurate diagnosis difficult. Therefore, in order to make an accurate diagnosis using an ultrasound image, it is necessary to remove or reduce speckle noise. The edge information in the ultrasound image has important information necessary for diagnosis visually or semantically. Therefore, it is important to ensure that the edge parts are not damaged during the noise cancellation process. To this end, techniques that reduce speckle noise and emphasize visually important parts such as edges are commonly used.

 Recently, a filtering technique based on multiscale wavelets has been used. It analyzes the edge of wavelet-decomposed image in pixel dimension and divides it into edge and speckle noise. This method is effective when applied to an ultrasound still image, but a problem arises when an ultrasound image sequence is targeted. When the method of processing the parts located at the same position in the adjacent frame is different, there is a problem that the flashing phenomenon occurs around the edge due to the local contrast difference of each frame in the image sequence. Therefore, in order to improve the quality of the B-mode ultrasound video, it is required to solve the above problem.

In order to solve the above problems, the present invention utilizes an edge adaptive filter based on wavelet transform and normalized edge map to enhance the edges, reduce the speckle noise, and reduce the glare around the edges. Provide ways to improve.

According to the present invention, there is provided a method of processing an ultrasound image, the method comprising: a) decomposing an ultrasound image by one level wavelet; b) extracting a weighted edge map using the eigenvalues of the tangential and normal directions in each pixel of the wavelet-decomposed image; c) performing steps a) to b) to the N level, wherein N is any positive integer; d) removing independent edges from the weighted edge map; e) adaptive filtering by applying different weights according to pixel values of each edge in the weighted edge map; f) wavelet synthesizing the adaptive filtered wavelet decomposition image; And g) repeating the steps e) to f) until the wavelet level of the wavelet decomposition image becomes 1.

1 is a flowchart illustrating a method of improving image quality of an ultrasound image according to the present invention. Referring to FIG. 1, a plurality of images having multiple resolutions are obtained through a one-level wavelet transform on an original ultrasound image (S110). Here, an original ultrasound image is passed through a low pass filter and downsampled to obtain an approximation coefficient. A weighted edge map is generated by extracting an edge using a structure tensor of an approximation image having the approximation coefficient thus obtained (S120). Wavelet transform and edge extraction are repeatedly performed until the wavelet level becomes N level (S130). Where N is any positive integer.

Hereinafter, a method of extracting an edge from an approximate image will be described in detail. In order to extract the edge information from the approximated image obtained by the wavelet transform, a structural matrix is obtained by using the gradients of all pixels in the approximated image. The eigenvalue decomposition of the structure matrix thus obtained calculates two eigenvalues at each pixel of the approximate image. Edge information is extracted from the image based on the calculated difference between two eigenvalues. Hereinafter, a process of extracting edge information using a structure tensor will be described in detail. For example, the structural tensor S _ρ of an arbitrary pixel a _j (x, y) in the wavelet-decomposed approximation image at j level may be expressed by Equation 1 below.

이고,

는 텐서곱(tensor product)을 나타내며,

는 컨볼류션(convolution)을 나타내고,

는 다음과 수학식2와 같이 정의되는 2차원 가우시안(Gaussian) 함수이다.From here,

ego,

Represents the tensor product,

Represents convolution,

Is a two-dimensional Gaussian function defined as in Equation 2 below.

에 대하여 고유값 분해를 하면 수학식3과 같은 식을 얻을 수 있다.Structural tensor

When eigenvalue decomposition is given, Equation 3 can be obtained.

Here, ν ₁ and ν ₂ are eigen vectors in a direction parallel to the edge and in a direction perpendicular to each other, and μ ₁ and μ ₂ each represent an eigen value that is the size of each eigenvector. These eigenvalues represent the average of the pixel values in the eigenvector direction.

)을 다음의 수학식4와 같이 정의할 수 있다.Since the difference in the eigenvalues is large in the edge region, the edge information of each pixel can be extracted using the eigenvalue information. A weighted edgemap that represents the probability that each pixel belongs to an edge in the wavelet-decomposed approximation image at j-level (

) Can be defined as in Equation 4 below.

Here, T _j represents a threshold for adjusting the ratio of the edge. In Equation 4, if | μ ₁ -μ ₂ | is large, the probability that the pixel belongs to the edge is high, and conversely, the closer | μ ₁ -μ ₂ | is to 0, the more speckle or homogeneous the pixel is. Highly likely to belong to the area. FIG. 2 illustrates a computationally weighted edge map as a 3D image in an actual ultrasound image. As shown in FIG. 2, it can be seen that there are various values between the edge and the homogeneous area.

Meanwhile, looking at the weighted edge map shown in FIG. 2, small edge components of independent shapes may be observed. Since these independent edge components act as the main cause of sparkling in the speckle or homogeneous area, it is necessary to lower the weight for the independent edge components. To this end, a low pass filter such as an intermediate value filter or an average value filter is applied (S140). 3 illustrates an image of a result of applying a median filter to a weighted edge map. In FIG. 3, a portion having a large pixel value is a portion determined to be a certain edge. It can be seen in FIG. 3 that the weights of the independent edge components are lowered while maintaining the boundaries of certain edge portions.

Subsequently, according to an embodiment of the present invention, an adaptive filter incorporating an edge enhancement filter and a speckle reduction filter is waveleted to an N level in order to alleviate edge enhancement, speckle reduction, and sparkling due to independent edges during video observation. By applying the weighted edge map to the decomposed image, the edges are made clear and the speckle noise is reduced (S150).

)에 가중되는 가중치(

)가 1에 가까울수록 그 픽셀는 에지에 포함되어 있을 확률이 높아지므로 이때에는 에지를 상대적으로 많이 강조하고 스페클 저감 효과를 작게 한다. 반대로, 가중치(

)가 0에 가까울수록 그 픽셀이 스페클 영역 또는 균질한 영역에 포함되어 있을 확률이 높아지므로 에지를 상대적으로 적게 강조하고 스페클 저감 필터의 영향력을 높인다.Hereinafter, the adaptive filtering method according to the present invention will be described in detail. The adaptive filter method according to the present invention provides an arbitrary pixel value of the weighted edge map (

Weighted to

The closer to) is, the higher the probability that the pixel is included in the edge, so that the edges are emphasized relatively much and the speckle reduction effect is small. Conversely, weights (

The closer to) is, the higher the probability that the pixel is in the speckle region or homogeneous region, thus emphasizing the edge less and increasing the impact of the speckle reduction filter.

As shown in Fig. 4, the eigenvector ν ₁ of the structural tensor is a vector of the edge and the normal direction and ν ₂ represents a vector of the edge and the tangential direction. If the size of both vectors is 1, the coordinates of the image in real implementation are not real numbers but real values, so they can be obtained using bidirectional linear interpolation using the values of integer points around them.

)에 대한 적응 필터링으로 나타낼 수 있다.4 shows a linear interpolation method of adjacent pixel values using eigenvectors and directionality at an edge. In FIG. 4, T ₁ and T ₂ are adjacent pixel values in the tangential direction to the edge, and N ₁ and N ₂ are adjacent pixel values in the edge and normal direction. Using these pixel values, the continuity can be enhanced by directional smoothing in the tangential direction of the boundary, and the boundary can be sharpened by directional sharpening in the normal direction. Approximation coefficient (such as

) Can be represented by adaptive filtering.

는 적응 필터링을 실시한 픽셀 값이며,

는 적응 필터링을 실시하기 전의 픽셀값이다. α는 방향성 평활화의 필터 계수로서

이고, β는 방향성 선명화의 필터 계수를 나타낸다. 여기서, k는 방향성 평활화의 정도를 조절하는 상수이다. 그리고 j레벨로 웨이블릿 변환과정에서 고역 통과 필터 및 다운샘플링을 통하여 얻은 상세 계수(detail coefficient)로 이루어진 상세 영상의 픽셀값(

)에 대해서는 수학식6과 같은 필터링 과정을 수행한다.The meaning of Equation 5 is to apply a low pass filter in the tangential direction and a high pass filter in the normal direction. here,

Is the pixel value with adaptive filtering.

Is the pixel value before performing adaptive filtering. α is the filter coefficient of the directional smoothing

Is a filter coefficient for directional sharpening. Where k is a constant for adjusting the degree of directional smoothing. The pixel value of the detailed image including the detail coefficients obtained through the high pass filter and the downsampling during the wavelet transform to the j level (

), The filtering process as in Equation 6 is performed.

,

이며, l₁ 및 l₂는 방향성 선명화를 결정하는 상수이고, λ_n은 웨이블릿 수축(shrinkage) 정도를 결정하는 상수이다. 상세 영상의 경우 에지 방향으로 평활화 필터를 적용하여 에지의 연속성을 높인다. 그러나

인 픽셀에서는 에지 선명화 필터의 영향으로 스페클 잡음의 영향이 상대적으로 커져서 오히려 잡음을 증가시킬 우려가 있기 때문에 l₁보다 작은 l₂를 적용하여 방향성 선명화의 정도를 감소시킨다. 스페클이 상세 영역에서 많이 관찰되므로 스페클에 해당되는 고주파 성분에 λ를 곱하여 수축시킴으로써 스페클 잡음을 줄일 수 있다. 수학식6에 나타나는 바와 같이, 스페클 저감 필터에서 웨이블릿 계수를 수축시키는 상수 λ가 0으로 가까워질수록 웨이블릿 계수가 지나치게 작아지기 때문에 특정한 값 이하에서는 수축 상수 λ를 일정하게 유지하도록 한다.From here,

,

L ₁ and l ₂ are constants for determining directional sharpening, and λ _n is a constant for determining the degree of wavelet shrinkage. In the case of a detailed image, a smoothing filter is applied in the edge direction to increase edge continuity. But

In in-pixel, the effect of the speckle noise is relatively increased due to the edge sharpening filter, which may increase the noise. Therefore, l ₂ smaller than l ₁ is applied to reduce the degree of directional sharpening. Since the speckle is observed in the detail region, the speckle noise can be reduced by multiplying the high frequency component corresponding to the speckle by λ. As shown in Equation 6, since the wavelet coefficient becomes too small as the constant lambda shrinking the wavelet coefficient approaches zero in the speckle reduction filter, the shrinkage constant lambda is kept constant below a specific value.

In the embodiment of the present invention, the weight of the filter is determined by the gamma function and the monotonic increase function. At this time, if k, l ₁ , l ₂ , and λ _n are fixed, the filter performance may be changed in a region where the edge probability is low without affecting the edge region (when W _j (x, y) = 1). Can be.

)는 수학식7과 같이 감마 함수(gamma function, f₁(x))와 단조 증가 함수(monotonic increase function, f₂(x))로 정의된다.The weight of the filtering (

) Is defined as a gamma function (f ₁ (x)) and a monotonic increase function (f ₂ (x)).

의 값에 따라 변할 수 있다.The output of the gamma function and monotonic increment function

It can vary depending on the value of.

값에 따른 감마 함수와 단조 증가 함수의 변화를 나타낸 그래프이다. 감마함수에서

가 1보다 커질수록 에지외의 영역에서 방향성 필터의 영향이 커지게 된다. 반대로 1보다 작은 경우에는 방향성 필터의 영향력이 감소하게 된다. 단조 증가 함수에서는

의 값을 0보다 큰 값을 줌으로써 스페클 영역에서의 방향성 필터의 영향을 크게 할 수 있다.5 is a graph showing conversion of weights by a gamma function and a linear increase function, respectively. In particular, Figure 5

It is a graph showing the change of gamma function and monotonic increase function according to the value. In gamma function

If is greater than 1, the influence of the directional filter in the region other than the edge becomes larger. On the contrary, if it is less than 1, the influence of the directional filter is reduced. In the monotonic increment function

By giving the value of 0 to a value greater than 0, the influence of the directional filter in the speckle region can be increased.

)에 따라서 필터의 특성이 변하게 된다.FIG. 6 is a graph showing normalized frequency response of a filter used in the tangential direction in directional filtering, and FIG. 7 is a graph showing the frequency response of the filter used in the normal direction in directional filtering. As can be seen in Figures 6 and 7, when α is constant, the weight (

), The filter characteristics change.

Subsequently, (l-1) the level image is synthesized from the N-level image subjected to the adaptive filtering (S160). This process is repeated until the wavelet level l becomes 1 (S170).

8 to 11 show the original images and the processed images one by one in the liver and fetal image sequences used in the experiment, respectively. Comparing the drawings, it can be seen that the edge portion of the processed image is clear and the noise component around the edge is suppressed. It can also be seen that the speckle noise in the homogeneous region is significantly suppressed. Also, if you look at the video, you can see that the edges are much less shiny.

While the present invention has been described and illustrated by way of preferred embodiments, those skilled in the art will recognize that various modifications and changes can be made without departing from the spirit and scope of the appended claims.

As described above, according to the present invention, an edge-adaptive filter can be used to enhance an edge and suppress noise components at the periphery, thereby obtaining an ultrasound image having improved image quality. In particular, when the image processing method according to the present invention is applied to an ultrasonic video, subjective improvement of image quality can be obtained by suppressing a sparkling phenomenon around the edge due to the local contrast difference of each frame.

Claims (8)

a) decomposing the ultrasound image by one level wavelet; b) extracting a weighted edge map using the eigenvalues of the tangential and normal directions in each pixel of the wavelet-decomposed image; c) performing steps a) to b) to the N level, where N is any positive integer; d) removing independent edges from the weighted edge map; e) adaptive filtering by applying different weights according to pixel values of each edge in the weighted edge map; f) wavelet synthesizing the adaptive filtered wavelet decomposition image; And g) repeating steps e) to f) until the wavelet level of the wavelet decomposition image is 1; Ultrasonic image processing method comprising a. The method of claim 1, B), b1) obtaining a structural tensor using gradients of each pixel of the decomposed image; b2) calculating eigenvalues in the tangential and normal directions at each pixel through eigenvalue decomposition for the structural tensor; And b3) If the absolute value of the eigenvalue difference in the tangential direction and the normal direction is greater than or equal to the threshold, the pixel value of the edge is defined as "1". Extracting the weighted edge map by defining an absolute value divided by the threshold value; Ultrasonic image processing method comprising a. The method of claim 2, Step d) is a method of processing the ultrasound image to remove the independent edge by applying a low pass filter to the edge extracted in step c). The method of claim 3, wherein And the low pass filter is an average filter or a median filter. The method of claim 2, And in step e), the weight is determined by a gamma function or a monotonically increasing function that inputs pixel values of each edge in the weighted edge map. The method of claim 5, wherein Step e), e1) performing a low pass filter in a tangential direction on the wavelet decomposition image by applying the determined weight value; And e2) applying a high pass filter in a normal direction to the wavelet decomposition image by applying the determined weight value Ultrasonic image processing method comprising a. The method of claim 6, In the step e), the ultrasound image processing method performing a process represented by the following equation. (Mathematical formula)

here,

Is the pixel value with adaptive filtering.

Is the pixel value before adaptive filtering, and α is the filter coefficient of directional smoothing,

Β is the filter coefficient of the directional sharpening

Are weights, k is a constant controlling the degree of directional smoothing, T ₁ , T ₂ are adjacent pixel values in the tangential direction to the edge, and N ₁ , N ₂ are pixel values adjacent in the direction normal to the edge. The method of claim 6, The step e) further comprises the step of reducing the speckle noise.

Images