JP6418922B2 - Image processing apparatus and image processing method - Google Patents

Description

  The present invention relates to an image processing technique that suitably reduces line noise in an image.

  In the medical field, diagnosis and treatment based on X-ray imaging are actively performed. Recently, an X-ray imaging apparatus using an FPD (Flat Panel detector) has been frequently used. The FPD realizes flattening of the X-ray detector by arranging photodiodes in a matrix, and the electrons generated in the conventional X-ray imaging apparatus using the image intensifier (II). Solves optical image distortion. In addition, since handling is easy due to the reduction in size and weight, it is possible to shoot with a high degree of freedom such as portable and mobile. However, since FPD reads a signal photoelectrically converted by a photodiode through a fine signal line, noise is likely to occur in an image due to the influence of external or internal factors. In addition, since FPD samples and holds signals for one row in the main scanning direction at a time and performs A / D conversion, line noise along the main scanning direction is likely to occur.

  In imaging using X-rays, imaging with a low dose is required in order to reduce the exposure of the human body. For this reason, the signal read by the X-ray detector has a very small value, and even if a slight fluctuation occurs in the image, it is visually recognized. In particular, line noise, which is streak-like unevenness, is sensitive to the human eye and is likely to affect diagnosis and treatment. Therefore, a technique using a spatial filter is known as a technique for reducing line noise. For example, Patent Document 1 discloses a method in which an input image including line noise is reduced in a direction parallel to the line noise, and the line noise extracted by filtering in the vertical direction is subtracted from the input image. It is disclosed. Further, Patent Document 2 discloses a line noise reduction method in which a pixel value-dependent filter such as an ε filter is used as a filter for extracting line noise to increase the ability to separate a subject and line noise, and artifacts are unlikely to occur. Has been. Furthermore, Non-Patent Document 1 describes an ε filter that is less affected by the structure of the subject by providing the ε filter with tilt adaptability.

Japanese Patent No. 4879938 Japanese Patent No. 5315157

Shinji Muneyasu and Ryo Taguchi, "Nonlinear Digital Signal Processing" Asakura Shoten 1999

  However, in the above-described method of Patent Document 1, separation of the subject and the line noise is performed by threshold processing, so that there is a possibility that the separation of the subject and the line noise is insufficient. In the method of Patent Document 2, when there is a subject structure, the difference between the filter target pixel value and the peripheral pixel value becomes large, and a sufficient line noise reduction effect may not be obtained. Furthermore, the method of Non-Patent Document 1 cannot distinguish between line noise and random noise, and there is a possibility that a sufficient effect of reducing line noise cannot be obtained.

  The present invention has been made in view of such problems, and an object of the present invention is to provide a technique that can more effectively reduce line noise in an image.

  In order to solve the above-described problems, an image processing apparatus according to the present invention has the following configuration. That is, the image processing apparatus performs a one-dimensional low-pass filter process on the input image obtained by capturing the subject in a direction parallel to the direction of the line noise included in the input image, thereby generating a random noise reduced image. A filtering means, a determining means for performing a local fitting process on the random noise reduced image along a direction orthogonal to the direction of the line noise, and determining as an image component of the subject, and the random noise reduced image Extraction means for performing a predetermined filter process along a direction orthogonal to the direction of the line noise to the image obtained by subtracting the image component of the subject from the image, and the input Generating means for subtracting the image component of the line noise from the image to generate a line noise reduced image.

  According to the present invention, it is possible to provide a technique that can more effectively reduce line noise in an image.

It is a figure which shows the function structure of the X-ray imaging system containing the image process part which concerns on 1st Embodiment. It is a flowchart which shows operation | movement of the image process part in 1st Embodiment. It is a figure which shows the flow of the image processing in an image processing part illustratively. FIG. 4 is a diagram exemplarily showing a pixel value profile of a pixel column including a subject structure. FIG. 4 is a diagram exemplarily showing a pixel value profile of a pixel column after subject structure removal. It is a figure which shows the function structure of the X-ray imaging system containing the image process part which concerns on 2nd Embodiment. It is a flowchart which shows operation | movement of the image process part in 2nd Embodiment.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. The following embodiments are merely examples, and are not intended to limit the scope of the present invention.

(First embodiment)
As a first embodiment of an image processing apparatus according to the present invention, an image processing unit included in an X-ray imaging system will be described below as an example.

<System configuration>
FIG. 1 is a diagram illustrating a functional configuration of an X-ray imaging system including an image processing unit according to the first embodiment.

  The X-ray tube 101 is a functional unit that irradiates the subject 103 with X-rays. The X-ray generator 104 is a functional unit that generates an X-ray by applying a high voltage pulse to the X-ray tube triggered by the user pressing the exposure switch. An FPD (Flat Panel Detector) 102 is a functional unit that is controlled by the FPD control unit 105, converts X-rays that have passed through the subject 103 into visible light using a phosphor, and detects them with a photodiode. The detected electrical signal is A / D converted and transmitted to the FPD control unit 105 as digital data.

  The FPD control unit 105 includes an image processing unit 109 and an image storage unit 108 and incorporates one or a plurality of computers. The computer includes, for example, a processing unit such as a CPU and a storage unit such as a ROM (Read Only Memory) and a RAM (Random Access Memory). Further, the computer may include a graphic control unit such as a GPU (Graphics Processing Unit), a communication unit such as a network card, an input / output unit such as a keyboard, a display, or a touch panel. These units are connected by a bus or the like, and are controlled by a processing unit such as a CPU executing a program stored in the storage unit.

  The monitor 106 is a functional unit that displays the received digital signal or the digital signal processed by the image processing unit 109 as an image. The operation unit 107 is a functional unit that receives an operation from the user and inputs an instruction to the image processing unit 109 or the FPD 102. The image storage unit 108 is a functional unit that stores the digital signal output from the FPD control unit 105 and the image data processed by the image processing unit 109.

  The image processing unit 109 includes a filter processing unit 110, a subject structure extraction unit 111, a pixel value dependent filter unit 112, a line noise extraction unit 113, and a line noise subtraction unit 114. The filter processing unit 110 generates a random noise reduced image from the input image. The subject structure extraction unit 111 extracts a subject structure from the input image. The pixel value-dependent filter unit 112 performs a pixel value-dependent high-pass filter process configured such that the filter coefficient changes depending on the pixel values of the pixels constituting the input image. The line noise extraction unit 113 extracts line noise using the subject structure extraction unit 111 and the pixel value dependent filter unit 112. The line noise subtractor 114 subtracts the line noise extracted by the line noise extractor 113 from the input image to generate an image with reduced line noise.

<Operation of Image Processing Unit>
FIG. 2 is a flowchart showing the operation of the image processing unit in the first embodiment. FIG. 3 is a diagram exemplarily showing the flow of image processing in the image processing unit.

  Under the control of the FPD control unit 105, signals (input images and X-ray captured images) detected and acquired by the FPD 102 are transferred to the image processing unit 109 via the bus. In this input image, as shown in FIG. 3A, image components of a subject 301, line noise 302, and random noise 303 such as Poisson noise and system noise are superimposed. Therefore, the image processing unit 109 performs processing for suitably removing the image component of the line noise 302 from such an input image. Note that as described in the background art, line noise is noise generated by sampling and holding a signal for one row in the main scanning direction in the FPD at a time and performing A / D conversion. For this reason, the line noise included in the input image is noise along one of the two pixel arrangement directions (X-axis direction and Y-axis direction) of the input image.

  In step S201, the filter processing unit 110 performs one-dimensional low-pass filter processing on the input image in a direction parallel to the line noise. This processing is intended to suppress Poisson noise caused by X-rays and system noise caused by the FPD 102 while maintaining the line noise. The one-dimensional low-pass filter process may use a known Gaussian filter, an average value filter, or the like. At this time, the input image may be reduced in the same manner as the methods described in Patent Document 1 and Patent Document 2 described above to speed up the subsequent processing.

  As a result of this processing, a random noise reduced image (FIG. 3B) in which Poisson noise and system noise are suppressed is obtained. The reason why the random noise reduced image is generated here is to allow line noise extraction described later to be executed more suitably.

  In step S202, the subject structure extraction unit 111 extracts an image component of the subject structure from the random noise reduced image obtained in S201. Specifically, by performing a local fitting process along a direction orthogonal to the direction of the line noise, an image component of a subject structure generally composed of low-frequency image components is extracted.

  FIG. 4 is a diagram exemplarily showing a pixel value profile of a pixel column including a subject structure. Specifically, the profile of the portion of the line segment 304 in FIG. Here, as shown in FIG. 3B, the line segment 304 is designated in a direction orthogonal to the direction of the line noise so as to cross the subject structure and the line noise. In this case, as indicated by a solid line 401, the profile of the line segment 304 is a profile in which an image component of line noise is superimposed on an image component of the subject structure.

  Since the subject structure is generally composed of low-frequency image components, it can often be fitted with a linear expression when viewed locally. In FIG. 4, the fitting result in the linear expression of the solid line 401 is a dotted line 402, and the object structure profile f (x, y + j) indicated by the dotted line 402 is expressed by the following mathematical expression (1).

  Here, x and y are the coordinates of the pixel that is the center of the fitting. I (x, y) is a pixel value of the random noise reduced image in FIG. 3B and corresponds to the solid line 401 in FIG. Further, y−n to y + n are fitting ranges.

  Note that step S202 is a process aimed at extracting the profile of the subject structure, and thus, for example, the profile f (x, y + j) of the subject structure may be extracted by performing low-pass filtering on the solid line 401. . In this case, the profile f (x, y + j) of the subject structure extracted by the subject structure extraction unit 111 is expressed by Equation (2).

  Here, G (K) is a low-pass filter coefficient. As will be described in the next step S203, G (K) in the subject structure extraction unit 111 and G (j) in the line noise extraction unit 113 may use the same filter coefficient. As a result, the subject structure can be extracted by limiting the low-frequency component below the cutoff frequency of the high-pass filter in the line noise extraction unit 113.

  In step S <b> 203, the line noise extraction unit 113 performs predetermined filter processing using the pixel value dependent filter unit 112 to extract line noise. Specifically, the line-shaped noise extraction unit 113 and the pixel value dependent filter unit 112 perform the process of Expression (3).

Here, G (J) is a coefficient of the low-pass filter, and is the same as G (K) in Expression (2). F (p) is a piecewise linear function. When the absolute value of p is smaller than ε, p is output as it is, and when the absolute value of p becomes ε or more, zero is output. That is, the filter coefficient changes in accordance with the difference between the filter target pixel value I (x, y) and the peripheral pixel value I (x, y + j) used for the filter.

  Hereinafter, step S203 will be described in detail. A general ε high-pass filter can be written in the form of Equation (4) below.

  Here, when ε is infinite, Expression (4) can be expressed as Expression (5) below.

  That is, it is understood that this corresponds to a high-pass filter obtained by subtracting the low-pass filter output from I (x, y). In Expression (4), when the difference between the pixel value to be filtered and the filter peripheral pixel value is larger than ε in Expression (5), the filter coefficient is set to zero to eliminate the influence of peripheral pixels that are strongly influenced by the subject structure. . Usually, the amplitude of the line noise is sufficiently smaller than that of the subject structure, and therefore, ε should be about the amplitude of the line noise.

  By the way, the method of Formula (4) can effectively extract line noise when the subject structure corresponds to an edge or a peak. However, in the case of the segment 304 in FIG. 3B, a profile in which the pixel value gradually changes depending on the subject structure is extracted as indicated by the solid line 401 in FIG. In this case, if it is attempted to extract line noise by the method of Equation (4), the difference from the filter target pixel becomes larger than the ε value in most peripheral pixels, and line noise cannot be extracted.

  Therefore, in the first embodiment, the line-shaped noise is more effectively extracted by extracting the profile f (x, y) of the subject structure and correcting I (x, y) as in Expression (3). . That is, in Equation (3), the processing of Equation (4) is performed after the subject structure f (x, y) obtained in Step S202 is removed from I (x, y).

  FIG. 5 is a diagram exemplarily showing a pixel value profile of a pixel column after subject structure removal. Specifically, the profile after the subject structure is removed from the portion of the line segment 304 in FIG.

  In this way, the image component derived from the subject structure is removed from the profile indicated by the solid line 401 in FIG. 4, and the profile indicated by the solid line 501 in FIG. 5 is generated. Then, the processing of Expression (4) is performed on the profile indicated by the solid line 501. Thereby, it is possible to preferably extract a line noise image as illustrated in FIG. 3C without being influenced by a gradual change in pixel value derived from the subject structure.

  Note that the influence of edges and peaks that cannot be extracted by subject structure extraction S202 can be removed by the piecewise linear function F. Therefore, Equation (3) is a pixel value-dependent filter that can extract line noise more suitably than the conventional technique. In the above description, an example using the ε filter has been described. However, a pixel value-dependent filter that depends on the relationship between the filter target pixel value and the peripheral pixel value, such as a bilateral filter or an MTM filter, can also be used. .

  In step S204, the line-shaped noise subtracting unit 114 configures the line-shaped noise image (FIG. 3C) extracted in S203 from the pixel values of the respective pixels configuring the input image (FIG. 3A). The pixel value of each pixel is subtracted. Thereby, a line-shaped noise reduced image as shown in FIG. 4D can be obtained.

  As described above, according to the first embodiment, it is possible to appropriately separate the image components of line noise by considering the subject structure included in the input image. As a result, it is possible to suitably remove an image component of line noise from the input image and generate a high-quality input image with reduced line noise.

(Second Embodiment)
In the second embodiment, a description will be given of a mode in which more accurate subject structure extraction is possible and, as a result, line noise can be reduced more appropriately. Specifically, the subject structure extraction is performed by a plurality of different processes, and a more accurate subject structure extraction result is adaptively selected.

<System configuration>
FIG. 6 is a diagram illustrating a functional configuration of an X-ray imaging system including an image processing unit according to the second embodiment. The schematic configuration of the entire system is the same as that of the first embodiment (FIG. 1). However, the image processing unit 609 is different from the first embodiment in that the image processing unit 609 further includes an extraction structure evaluation unit 615 that evaluates the extracted structure and a subject structure selection unit 616 that selects a subject structure according to the evaluation result.

<Operation of Image Processing Unit>
FIG. 7 is a flowchart showing the operation of the image processing unit in the second embodiment.

  A signal (input image) detected by the FPD 102 by the control of the FPD control unit 105 is transferred to the image processing unit 609 via the bus. In this input image, as shown in FIG. 3A, image components of a subject 301, line noise 302, and random noise 303 such as Poisson noise and system noise are superimposed. Therefore, the image processing unit 609 suitably performs processing for removing the image component of the line noise 302 from such an input image.

  In step S601, the filter processing unit 110 performs one-dimensional low-pass filter processing on the input image in a direction parallel to the line noise. Since this process is the same as that of the first embodiment, detailed description thereof is omitted.

  In step S602, the subject structure extraction unit 111 extracts an image component of the subject structure from the random noise reduced image obtained in S601. However, in the second embodiment, two or more different structure extraction methods are executed. Here, as an example, an example will be described in which the two structure extraction methods of Equation (1) and Equation (2) described above are executed.

The subject structure (first image component candidate) extracted by the linear expression of Expression (1) is derived as f ₁ (x, y + j). Further, the subject structure (second image component candidate) extracted by the filter of Expression (2) is derived as f _G (x, y + j).

  In addition, although the example which performs the two structure extraction methods of Numerical formula (1) and Numerical formula (2) is demonstrated here, you may utilize another structure extraction method. For example, the subject structure may be extracted using a quadratic or higher order polynomial, a sigmoid function, a trigonometric function, or the like instead of the linear expression such as Expression (1) and Expression (2). Moreover, you may comprise so that three or more different structure extraction methods may be performed.

  In step S603, the extraction structure evaluation unit 615 evaluates the accuracy (fitting accuracy) of the extraction structure with respect to the extraction results of the plurality of subject structures obtained in S602. As the evaluation function, for example, a residual sum of squares can be used as in the following formula (6).

  In addition, you may use a residual absolute value sum, a correlation coefficient, a determination coefficient, etc. as an evaluation function. Thereafter, the subject structure selection unit 616 selects a more appropriate subject structure extraction result based on the evaluation result of each subject structure by the evaluation function. For example, when the residual sum of squares is used as the evaluation function, it can be said that the subject structure is suitably extracted when the value becomes smaller.

By the way, since the subject structure is generally complicated, what kind of subject structure extraction method is appropriate depends on the position (x, y) of the image and the photographed subject. Therefore, in the second embodiment, in order to more appropriately extract the subject structure, the subject structure selection unit 616 adaptively outputs the subject structure extraction result that best represents the subject structure according to the position (x, y) of the image. select. The profile of the subject structure selected by the subject structure selection unit 616 is assumed to be f _S (x, y + j). Note that f _S (x, y + j) is obtained by selecting either f ₁ (x, y + j) or f _G (x, y + j) for each region.

  In step S <b> 604, the line noise extraction unit 113 extracts line noise using the pixel value dependent filter unit 112. Specifically, the line-shaped noise extraction unit 113 and the pixel value dependent filter unit 112 perform the processing of Expression (7) on the subject structure extraction result selected by the subject structure selection unit 616.

  By this processing, the line noise image shown in FIG. 3C can be obtained.

  In step S605, the line-shaped noise subtracting unit 114 configures the line-shaped noise image (FIG. 3C) extracted in S203 from the pixel values of the respective pixels configuring the input image (FIG. 3A). The pixel value of each pixel is subtracted. Thereby, a line-shaped noise reduced image as shown in FIG. 4D can be obtained.

  As described above, according to the second embodiment, the subject structure included in the input image can be extracted with higher accuracy, and the image components of the line noise are more preferably separated than in the first embodiment. It becomes possible to do. As a result, it is possible to suitably remove an image component of line noise from the input image and generate a high-quality input image with reduced line noise.

(Other examples)
The present invention supplies a program that realizes one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium, and one or more processors in a computer of the system or apparatus read and execute the program This process can be realized. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

  109 image processing unit; 110 filter processing unit; 111 subject structure extraction unit; 112 pixel value-dependent filter unit; 113 line noise extraction unit; 114 line noise subtraction unit

Claims (11)

First filter means for performing a one-dimensional low-pass filter process on the input image in which the subject is imaged in a direction parallel to the direction of the line noise included in the input image to generate a random noise reduced image;
A determination unit that performs local fitting processing along a direction orthogonal to the direction of the line noise on the random noise reduced image, and determines a fitting result as an image component of the subject;
Extracting the image obtained by subtracting the image component of the subject from the random noise-reduced image, performing predetermined filter processing along a direction orthogonal to the direction of the line noise, and extracting the image component of the line noise Means,
Generating means for subtracting the image component of the line noise from the input image to generate a line noise reduced image;
An image processing apparatus comprising: 2. The image according to claim 1, wherein the extraction unit changes a filter coefficient of the predetermined filter processing based on a pixel value of an image obtained by subtracting an image component of the subject from the random noise reduced image. Processing equipment. The image processing apparatus according to claim 1, wherein the predetermined filter process includes an ε filter, an MTM filter, or a bilateral filter. The determining means includes
First derivation means for deriving a first image component candidate based on a first local fitting process;
Second derivation means for deriving a second image component candidate based on a second local fitting process different from the first local fitting process;
Evaluation that compares each of the first image component candidate and the second image component candidate with the random noise reduced image and evaluates the fitting accuracy of the first image component candidate and the second image component candidate. Means,
Selection means for selecting one of the first image component candidate and the second image component candidate as the image component of the subject based on the evaluation result by the evaluation means;
The image processing apparatus according to claim 1, further comprising: The selection means selects any one of the first image component candidate and the second image component candidate for each of a plurality of regions constituting the input image as the image component of the subject. The image processing apparatus according to claim 4. The evaluation means includes, for each of the first image component candidate and the second image component candidate, a residual sum of squares, a residual absolute value sum, a correlation coefficient, and a determination coefficient with the random noise reduced image. The image processing apparatus according to claim 4, wherein at least one of the following is derived. The image processing apparatus according to claim 1, wherein the local fitting process includes a fitting process using a polynomial, a sigmoid function, a trigonometric function, or a low-pass filter. The image processing apparatus according to claim 7, wherein a filter coefficient of the low-pass filter is determined according to a filter coefficient of the predetermined filter process. The image processing apparatus according to claim 1, wherein the input image is an X-ray captured image acquired by an FPD (Flat Panel detector). An image processing method comprising:
A first filtering step of performing a one-dimensional low-pass filter process on the input image obtained by capturing the subject in a direction parallel to the direction of the line noise included in the input image to generate a random noise reduced image;
A determination step of performing a local fitting process along a direction orthogonal to the direction of the line noise on the random noise reduced image, and determining a fitting result as an image component of the subject;
Extracting the image obtained by subtracting the image component of the subject from the random noise-reduced image, performing predetermined filter processing along a direction orthogonal to the direction of the line noise, and extracting the image component of the line noise Process,
Subtracting the image component of the line noise from the input image, generating a line noise reduced image,
An image processing method comprising:   A program for causing a computer to function as each unit of the image processing apparatus according to any one of claims 1 to 9.

Images