JP5416377B2 - Image processing apparatus, X-ray foreign object detection apparatus including the same, and image processing method - Google Patents

Description

  The present invention relates to an image processing apparatus that processes, for example, an X-ray image, an X-ray foreign object detection apparatus that detects the presence or absence of a mixed foreign object based on the X-ray image, and an image processing method.

  Conventionally, as a device for processing an X-ray image, a restoration process using a point spread function (PSF) as an image restoration parameter causes the outline of the subject image to become unclear (hereinafter referred to as “image blur”). ) Is known to perform image restoration (for example, see Patent Document 1).

Patent Document 1 describes a method that divides image data that has been subjected to Fast Fourier Transform (FFT) processing by PSF that has been subjected to FFT processing, and performs deconvolution (deconvolution integration) by a method that performs inverse FFT processing on the calculation result. An operation is performed to generate a restored image.
JP 2008-73318 A

  However, it is known that the conventional one has a side effect that high-frequency noise is generated in the image blur restoration process by the deconvolution operation after the FFT process. This high-frequency noise can be suppressed by using, for example, a Wiener filter. However, if a Wiener filter is used, the blur image restoration effect is also suppressed by the low-pass filter effect, and minute foreign matter mixed in the object to be measured. There has been a problem that cannot be detected.

  This will be specifically described with reference to FIG. FIG. 12A shows in 3D the density of the original image of the object to be measured in which minute foreign objects exist in two circles. This original image shows a state in which a signal indicating a minute foreign object is buried in Poisson noise. FIG. 12B shows a restored image obtained by performing image restoration processing on the original image using a Wiener filter. As shown in FIG. 12B, it can be seen that minute foreign objects cannot be detected by the image restoration process using the Wiener filter.

  Further, in the conventional device, image blur due to X-ray scattering (hereinafter referred to as “scattering blur”) occurs, and a signal indicating a minute foreign matter is buried in Poisson noise, which simply improves the resolution of the X-ray detector. There is a problem that a minute foreign object cannot be detected by itself.

  In addition, an iterative method that performs iterative calculation for image restoration on an X-ray image is also known. However, in order to obtain sufficient image quality by this method, a large amount of processing time is required for performing iterative calculation. There has been a problem that it cannot be adopted in an X-ray foreign matter detection apparatus that requires processing.

  The present invention has been made in view of such circumstances, and an image processing apparatus capable of acquiring a high-definition image at a higher speed than conventional ones, an X-ray foreign object detection apparatus including the same, and image processing It aims to provide a method.

An image processing apparatus according to claim 1 of the present invention includes an image data input unit that inputs data of an object image including edge information indicating an edge of an object to be detected in an image obtained by photographing the object to be measured, and the detection object A function approximate point spread function generating means for generating a function approximate point spread function by approximating a point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard specimen of an object, and A restored image generating means for generating a restored image by performing an image restoration process by a deconvolution operation based on the function approximate point spread function based on a Gaussian function as a measured object image, and the edge information and a edge preserving smoothing means for removing noise from the restored image with saved state, said edge preserving smoothing means, the And a first edge preserving smoothing unit that performs a second edge preserving smoothing process, respectively, and the first edge preserving smoothing unit is a pixel of interest as the first edge preserving smoothing process. When the standard deviation of the local density including the neighboring pixels is equal to or less than the first threshold value, the image data having the simple average of the local density as the density of the target pixel is output to the second edge storage smoothing unit, and the local density When the standard deviation of the pixel exceeds the first threshold value, the simple average of the local density is corrected with a predetermined correction function, and the image data having the density of the target pixel is output to the second edge storage smoothing unit And the second edge preserving / smoothing unit, as the second edge preserving / smoothing processing, has a density of an image related to the image data output from the first edge preserving / smoothing unit with respect to the measured object image. Change value When the second threshold value is exceeded, the image data subjected to the edge preserving smoothing process by the bilateral filter is output, and when it is equal to or less than the second threshold value, the image data from the first edge preserving smoothing unit is output to the edge preserving smoothing process. The edge preserving smoothing means has a configuration in which the first and second edge preserving smoothing processes are sequentially performed twice each .

  With this configuration, the image processing apparatus according to claim 1 of the present invention generates a function approximate point spread function approximating the point spread function, performs image restoration by deconvolution operation, and stores edge information. Perform edge preserving smoothing to remove noise.

Further, this configuration, the image processing apparatus according to claim 1 of the present invention performs the edge preserving smoothing processing in the bilateral filter for only the edge portion which has not been smoothed in the first edge preserving smoothing process.

The image processing apparatus according to claim 2 of the present invention has a configuration in which the edge preserving smoothing unit includes an edge preserving smoothing unit that performs edge preserving smoothing processing a plurality of times using a bilateral filter.

With this configuration, the image processing apparatus according to claim 2 of the present invention performs the edge preserving smoothing process a plurality of times after the image restoration process by the deconvolution operation.

An image processing apparatus according to a third aspect of the present invention includes an image data input unit that inputs data of an object to be measured including edge information indicating an edge of the object to be detected in an image obtained by photographing the object to be measured; A function approximate point spread function generating means for generating a function approximate point spread function by approximating a point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard specimen of an object, and Edge-preserving smoothing means having first and second edge-preserving smoothing sections for performing first and second edge-preserving smoothing processes for removing noise in a state where edge information is preserved, and the edge-preserving smoothing An image related to the image data output by the means is a convolution image based on a Gaussian function, and a deconvolution based on the function approximate point spread function Restored image generating means for generating a restored image by performing an image restoration process by a calculation of the image, and the first edge preserving smoothing unit is a neighboring pixel of the target pixel as the first edge preserving smoothing process. When the standard deviation of the local density including is less than or equal to the first threshold, image data with the simple average of the local density as the density of the target pixel is output to the second edge storage smoothing unit, and the standard deviation of the local density When the image data exceeds the first threshold value, the simple average of the local density is corrected with a predetermined correction function, and the image data set as the density of the target pixel is output to the second edge storage smoothing unit, In the second edge preserving smoothing unit, as the second edge preserving smoothing process, a density change value with respect to the measurement object image of an image related to the image data output from the first edge preserving smoothing unit is obtained. The second When the threshold value is exceeded, the image data subjected to the edge preserving smoothing process by the bilateral filter is output, and when it is equal to or less than the second threshold value, the image data from the first edge preserving smoothing unit is not subjected to the edge preserving smoothing process. And the edge preserving smoothing means performs the first and second edge preserving smoothing processes twice in sequence with the image restoration process in between. doing.

With this configuration, the image processing apparatus according to claim 3 of the present invention performs the first edge preserving smoothing process, and then performs the image restoration process by the deconvolution operation to generate the restored image, and then the second Perform edge preserving smoothing.

In the image processing device according to claim 4 of the present invention, the bilateral filter includes a super Gaussian function determined by a parameter based on a distance between the target pixel and a neighboring pixel, and a density difference between the target pixel and the neighboring pixel. The weighting is set based on the sub-Gaussian function determined by the parameter based on the above, and the configuration is a double weighted moving average filter.

With this configuration, the image processing apparatus according to claim 4 of the present invention performs the edge preserving smoothing process using the bilateral filter as a double weighted moving average filter.

In the image processing apparatus according to claim 5 of the present invention, the Gaussian function includes a sub-Gaussian function and a super-Gaussian function, and the function approximate point spread function generation means is any one of the sub-Gaussian function and the super-Gaussian function. The function approximate point spread function is generated by determining one by the parameters.

With this configuration, the image processing apparatus according to claim 5 of the present invention generates a function approximate point spread function by determining any one of the sub-Gaussian function and the super-Gaussian function by a parameter.

According to a sixth aspect of the present invention, there is provided an X-ray foreign object detection device, an image processing device, an X-ray source that irradiates the object to be measured with X-rays, and an X-ray transmitted through the object to be detected. X-ray detection means for outputting to the image processing apparatus, and the image processing apparatus comprises a foreign substance detection means for detecting a foreign substance contained in the object to be measured based on the detection information. Yes.

With this configuration, the X-ray foreign object detection device according to claim 6 of the present invention generates a function approximate point spread function approximating the point spread function, performs image restoration by deconvolution operation, and stores edge information. An image processing apparatus that performs edge preserving smoothing processing for removing noise in a state is provided, and foreign matter contained in the object to be measured is detected.

An image processing method according to claim 7 of the present invention includes an image data input step of inputting data of a measured object image including edge information indicating an edge of the detected object in an image obtained by photographing the measured object; A function approximate point spread function generating step for generating a function approximate point spread function by approximating the point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard specimen of an object; An image restoration step for generating a restored image by performing an image restoration process by a deconvolution operation based on the function approximate point spread function based on a Gaussian function as a measured object image, and storing the edge information the look including an edge preserving smoothing step to remove noise from the restored image, the edge preserving smoothing in state The step includes first and second edge preserving smoothing steps for performing first and second edge preserving smoothing processes, respectively. In the first edge preserving smoothing step, the first edge preserving smoothing is performed. As processing, when the standard deviation of the local density including the neighboring pixels of the target pixel is equal to or smaller than the first threshold value, image data in which the simple average of the local density is set as the density of the target pixel is output, and the standard deviation of the local density is When the first threshold value is exceeded, a simple average of the local densities is corrected with a predetermined correction function to output image data having the density of the pixel of interest. In the second edge preserving smoothing step, As the edge preserving smoothing process, the density change value of the image relating to the image data output in the first edge preserving smoothing step with respect to the measured object image exceeds a second threshold value. Output image data that has undergone edge preserving smoothing processing using a bilateral filter, and output the image data from the first edge preserving smoothing step without performing the edge preserving smoothing processing when it is equal to or less than the second threshold value. In the edge preserving smoothing step, each of the first and second edge preserving smoothing processes is sequentially performed twice .

With this configuration, the image processing method according to claim 7 of the present invention generates a function approximate point spread function approximating the point spread function, performs image restoration by deconvolution operation, and stores edge information. Perform edge preserving smoothing to remove noise.

With this configuration, the image processing method according to claim 7 of the present invention performs the edge preserving smoothing process with the bilateral filter only for the edge portion that has not been smoothed in the first edge preserving smoothing process.

The image processing method according to an eighth aspect of the present invention has a configuration in which the edge preserving smoothing step includes an edge preserving smoothing step in which the edge preserving smoothing process is performed a plurality of times by a bilateral filter.

With this configuration, the image processing method according to claim 8 of the present invention performs the edge preserving smoothing process a plurality of times.

An image processing method according to claim 9 of the present invention includes an image data input step of inputting data of a measurement object image including edge information indicating an edge of the detection object in an image obtained by photographing the measurement object; A function approximate point spread function generating step for generating a function approximate point spread function by approximating the point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard specimen of an object; Edge-preserving smoothing step having first and second edge-preserving smoothing steps for performing first and second edge-preserving smoothing processes for removing noise in a state where edge information is preserved, respectively, and the edge-preserving smoothing The image related to the image data output in step is a convolution image based on a Gaussian function, and the function approximate point spread function is And an image restoration step of generating a restored image by performing an image restoration process by a deconvolution operation, and in the first edge preservation smoothing step, as the first edge preservation smoothing process, When the standard deviation of the local density including the pixel is less than or equal to a first threshold value, image data is output with a simple average of the local density as the density of the pixel of interest, and when the standard deviation of the local density exceeds the first threshold value A simple average of the local density is corrected with a predetermined correction function to output image data having the density of the target pixel. In the second edge storage smoothing step, as the second edge storage smoothing process, When the density change value of the image related to the image data output in the first edge preserving smoothing step with respect to the measured object image exceeds a second threshold value, Outputting image data subjected to edge preserving smoothing processing by a lateral filter, and outputting the image data from the first edge preserving smoothing step without performing the edge preserving smoothing processing when the second threshold value or less, In the edge preserving / smoothing step, the first and second edge preserving / smoothing processes are sequentially performed twice each with the image restoration step in between.

With this configuration, in the image processing method according to claim 9 of the present invention, after performing the first edge-preserving smoothing process, the image restoration process is performed by the deconvolution operation, and then the second image is generated. Perform edge preserving smoothing.

  An image processing apparatus according to claim 1 of the present invention generates a function approximate point spread function approximating a point spread function, performs image restoration by deconvolution operation, and removes noise while preserving edge information. By performing the edge preserving smoothing process, it is possible to obtain a higher-definition image at a higher speed than the conventional one.

The image processing apparatus according to claim 1 of the present invention, by performing edge preserving smoothing processing in the bilateral filter for only the edge portion which has not been smoothed in the first edge preserving smoothing process, than the conventional High-definition images can be acquired at high speed.

The image processing apparatus according to claim 2 of the present invention obtains a remarkably high-definition image at a higher speed than the conventional one by performing the edge preserving smoothing process a plurality of times after the image restoration process by the deconvolution operation. Can do.

An image processing apparatus according to a third aspect of the present invention performs a first edge preserving smoothing process, then performs an image restoration process by a deconvolution operation to generate a restored image, and then a second edge preserving smoothing By performing the processing, it is possible to obtain a higher-definition image than the conventional one at high speed.

The image processing apparatus according to claim 4 of the present invention can acquire a finer image at a higher speed than the conventional one by performing edge preserving smoothing processing using a bilateral filter as a bi-weighted moving average filter. .

The image processing apparatus according to claim 5 of the present invention generates a function approximate point spread function by determining any one of a sub-Gaussian function and a super-Gaussian function by a parameter, thereby generating a function approximate point spread function. High-definition images can be acquired at high speed.

According to a sixth aspect of the present invention, an X-ray foreign object detection device generates a function approximate point spread function approximating a point spread function, performs image restoration by deconvolution operation, and generates noise in a state where edge information is stored. Equipped with an image processing device that performs edge-preserving smoothing processing to remove, and by detecting foreign matter contained in the object to be measured, scattering blur occurs and is buried in Poisson noise. It can be detected in real time.

An image processing method according to claim 7 of the present invention generates a function approximate point spread function approximating a point spread function, performs image restoration by deconvolution operation, and removes noise while preserving edge information. By performing the edge preserving smoothing process, it is possible to obtain a higher-definition image at a higher speed than the conventional one.

The image processing method according to claim 7 of the present invention performs edge preserving smoothing processing with a bilateral filter only for the edge portion that has not been smoothed in the first edge preserving smoothing processing, so that the conventional image processing method can be used. High-definition images can be acquired at high speed.

In the image processing method according to the eighth aspect of the present invention, by performing the edge preserving smoothing process a plurality of times, it is possible to obtain a higher-definition image than the conventional one at high speed.

In the image processing method according to claim 9 of the present invention, after performing the first edge preserving smoothing process, an image restoration process is performed by a deconvolution operation to generate a restored image, and then the second edge preserving smoothing is performed. By performing the processing, it is possible to obtain a higher-definition image than the conventional one at high speed.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(First embodiment)
First, the configuration of the X-ray foreign object detection device in the first embodiment of the present invention will be described.

  As shown in FIG. 1, the X-ray foreign object detection device 10 in this embodiment includes an X-ray source 1 that irradiates X-rays, a measurement object 2 that is irradiated with X-rays, and a conveyance that conveys the measurement object 2. Means 3, an X-ray detector 4 for detecting X-rays transmitted through the object to be measured 2 and the conveying means 3, an image processing device 5 for performing image processing, an inspection information input means 6 for inputting inspection information, and an image The image display means 7 is displayed.

  The X-ray source 1 has an X-ray tube that generates X-rays by colliding a thermoelectron from a cathode filament with an anode target by a high voltage between the cathode and the anode, for example. The beam is shaped and irradiated toward the lower X-ray detector 4 by a slit (not shown) shaped into a fan-shaped beam extending in the width direction of the conveying means 3 (direction perpendicular to the conveying direction). That is, the X-ray source 1 and the X-ray detector 4 constitute a so-called X-ray fan beam optical system.

  The object to be measured 2 is, for example, food, medicine, or the like, and is a detection target for contained fine foreign matter. In FIG. 1, the device under test 2 is shown as an individual for the sake of convenience, but the device under test 2 may be in the form of a rose or powder containing granular materials, flakes, powders, and the like.

  The conveyance means 3 is constituted by, for example, a belt conveyor, and conveys the object 2 to be measured at a predetermined constant conveyance speed corresponding to the product type, and the object 2 is measured in the middle of the conveyance with the X-ray source 1 and X-ray detector. 4 is passed through. The X-rays irradiated from the X-ray source 1 are absorbed by the object to be measured 2 and slightly absorbed by the belt conveyor, pass through them, and reach the X-ray detector 4.

  The X-ray detector 4 detects X-rays that have passed through the DUT 2 for each of a plurality of transmission areas adjacent in the width direction of the transport means 3, and stores data on the accumulated transmission amount for each predetermined time in each transmission area. It is output as detection information. The X-ray detector 4 is configured by, for example, an X-ray line sensor, although details are not shown. This X-ray line sensor is a publicly known sensor in which detection elements are arranged in an array in the width direction of the transport means 3 at a predetermined pitch, and can detect X-rays with a predetermined resolution.

  The aforementioned X-ray source 1, transport means 3 and X-ray detector 4 are each controlled by a control device (not shown) so as to operate at a predetermined timing. This control device is configured to include a microcomputer having, for example, a CPU, a ROM, a RAM, and an I / O interface, and executes conveyance and X-ray irradiation / detection control according to a predetermined program stored in the ROM. It has the program etc.

  The image processing apparatus 5 includes an image input unit 51 that inputs image data, an image restoration unit 52 that restores an image, a foreign matter detection unit 53 that detects a foreign matter, and a foreign matter determination unit 54 that determines whether or not a foreign matter is present. I have. The image processing device 5 has the same configuration as that of the above-described control device, and the microcomputer executes an image processing program for detecting minute foreign matter.

  The image input means 51 A / D-converts X-ray detection signals from a plurality of detection elements of the X-ray detector 4 and conveys them at every predetermined unit conveyance time corresponding to the arrangement pitch of the detection elements. For all n (n is a positive integer, for example, 640) transmission region in the width direction, data of accumulated transmission X-ray dose (hereinafter referred to as “transmission amount”) within the unit time is obtained from 0 to 1023, for example. It can be output to the outside as transmission detection level line detection information representing gradation, and has an A / D converter, a memory, and the like (all not shown). Note that the image input means 51 has no X-ray transmission amount corresponding to the transmission amount data written in the memory of the image input means 51 when line scanning is performed on each DUT 2. When the value of X is the maximum and the X-ray absorption amount by the object to be measured 2 becomes zero, the minimum density value is obtained, and when the value of the X-ray transmission amount is the minimum and the X-ray absorption amount by the object to be measured 2 becomes maximum An X-ray image converted to equivalent thickness data using logarithmic conversion so as to obtain a density value is output. The image input means 51 constitutes image data input means according to the present invention.

  The image restoration unit 52 includes a parameter setting unit 521, a PSF generation unit 522, a deconvolution unit 523, and an edge preserving smoothing unit 524, and an X-ray image (hereinafter referred to as “measurement object image”) input from the image input unit 51. ) Data is subjected to image restoration processing including local processing using a kernel of kernel size N (N × N pixels).

  The parameter setting unit 521 includes, for example, an input device such as a keyboard and a mouse, a memory for storing parameter data, and the like. This parameter is an edge-preserving smoothing parameter, for example, for weighting the pixel values of neighboring pixels according to the density and distance centered on a pixel whose pixel density value is to be obtained in the kernel (hereinafter referred to as “target pixel”). Parameters (described later) and the like, and includes a kernel size N, a standard deviation σ, and an index γ for setting a function approximation PSF (described later). Before performing the foreign substance detection process, the data of these parameters are obtained when the standard test piece of the minute foreign matter to be detected is included in the object 2 and the standard test piece of the fine foreign object is imaged most clearly. The data is adopted. As the standard test piece, for example, a metal sphere having a diameter of about 0.2 mm to 2 mm is preferably used. The parameter data is preferably set according to the X-ray energy value irradiated by the X-ray source 1, the element size of the X-ray detector 4, the type of the X-ray detector 4, the type of the object to be measured, and the like. . The parameter data is preferably determined in consideration of the type and size of the minute foreign matter.

  The PSF generation unit 522 generates a function approximate PSF that approximates the PSF with a parametric function based on the parameters σ and γ set by the parameter setting unit 521, and outputs the data to the deconvolution unit 523 and the edge preserving smoothing unit 524. It is supposed to be. The PSF generation unit 522 constitutes a function approximate point spread function generation unit according to the present invention.

  Specifically, the PSF generation unit 522 generates a function approximate PSF based on σ and γ set by the parameter setting unit 521 by using equation (1). R indicates the distance from the target pixel to the predetermined pixel.

PSF (r) ∝ exp {-0.5 ・ (r / σ) ^γ } (1)

  Here, the function approximation PSF is a general Gaussian function if γ = 2, a sub-Gaussian function if γ> 2, and a super Gaussian function if γ <2, and FIG. 2 shows a perspective view of the approximation model. FIG. 2A shows a function approximation PSF model (left figure) obtained by function approximation with N = 11, σ = 1.2, and γ = 2.5 (sub-Gaussian), and a PSF function model (right figure) in its analog image. Figure). FIG. 2B shows a function approximation PSF model (left figure) obtained by function approximation with N = 11, σ = 1.2, and γ = 1.5 (super Gaussian), and a PSF function model in an analog image thereof. (Right figure). Here, it is desirable that the kernel size N is a minimum size including the entire function approximate PSF model (≠ 0). As shown in FIG. 2, the processing speed can be increased by modeling the PSF by function approximation and simplifying the pixel value weighting calculation.

  Returning to FIG. 1, the deconvolution means 523 includes, for example, a Wiener filter and an inverse filter (Inverse Filter), the measured object image is a convolution image based on a Gaussian function (σ, γ), and is based on a function approximation PSF. Thus, a restored image is generated by performing an image restoration process by a deconvolution operation. Note that the deconvolution means 523 constitutes a restored image generation means according to the present invention.

  The edge preserving / smoothing means 524 includes a first edge preserving / smoothing unit 524a and a second edge preserving / smoothing unit 524b that perform first and second edge preserving smoothing processes, respectively. Further, the edge preserving / smoothing means 524 includes a memory (not shown) for storing each data of the first threshold value and the second threshold value acquired in advance by including a standard test piece of minute foreign matter in the DUT 2. A standard test piece of a foreign object that uses the first threshold value and the second threshold value when the image is most clearly imaged is stored. As will be described later, the edge preserving / smoothing means 524 performs the first and second edge preserving / smoothing processes twice in sequence.

  The first edge preserving / smoothing unit 524a includes, for example, an adaptive moving average filter, inputs restored image data from the deconvolution means 523, and calculates an average value image obtained by equally weighting the inside of a square window having a kernel size N ′. At the same time, the standard deviation S of the density values in the square window is obtained. Further, when the first threshold value is represented by A, the first edge preserving / smoothing unit 524a outputs the data of the average value image if the standard deviation S ≦ the first threshold value A. That is, the first edge preserving / smoothing unit 524a increases the processing speed by outputting average image data for an image that does not include an edge.

Further, the first edge preserving / smoothing unit 524a outputs the image data P obtained by the correction shown in the equation (2) if the standard deviation S> the first threshold A. In the equation (2), I represents the pixel density of the measured object image, and I ′ represents the pixel density of the average value image. Further, rate = (S ² −A ² ) / (S ² + A ² ).

  P = I '+ (I-I') x rate (2)

  The second edge preserving / smoothing unit 524b includes, for example, a bilateral filter, and outputs image data based on the density change value of the image after the first edge preserving smoothing process with respect to the measured object image. It is like that.

  Here, the Gaussian function is expressed as shown in the equation (3) using the distance r from the center, the standard deviation σ, and the index γ as parameters.

G (r, σ, γ) = (0.4 / σ) · exp {−0.5 (r / σ) ^γ } (3)

  When the second threshold value is represented by B and the density change value> the second threshold value B, the bilateral filter calculates σ ′ and γ ′ with respect to the distance Δd between the target pixel and the neighboring pixel as shown in the equation (4). Movement weighted with a bi-weighting coefficient W set based on a super Gaussian function as a parameter and a sub-Gaussian function with σ ″ and γ ″ as parameters for the density difference Δi between the target pixel and neighboring pixels An average filter, which outputs data of an average value image weighted by a bi-weighting coefficient W. In this case, the bilateral filter functions as a dual weighted moving average filter.

W ∝ exp {−0.5 · (Δd / σ ′) ^{γ ′} } · exp {−0.5 · (Δi / σ ″) ^{γ ″} } (4)

  Specifically, for example, a super Gaussian function indicated by G (Δd, σ ′, 1) and a sub-Gaussian function indicated by G (Δi, σ ″, 4) in a round window having a kernel size N ″. And the bi-weighting coefficient W is obtained.

  On the other hand, the second edge preserving / smoothing unit 524b outputs the image data after the first edge preserving / smoothing process as it is when the density change value ≦ the second threshold value B.

  As described above, the second edge preserving / smoothing unit 524b uses the bilateral filter as an edge only for the edge portion (density change value> second threshold B) that has not been smoothed by the first edge preserving smoothing unit 524a. A storage smoothing process is performed.

  The foreign matter detection means 53 detects the fine foreign matter mixed in the measurement object together with the non-fine foreign matter based on the image restored by the image restoration means 52, and outputs the foreign matter image data indicating the foreign matter to the foreign matter determination means 54. It is like that.

  The foreign matter determination means 54 determines the minute foreign matter mixed in the DUT 2 based on the foreign matter image data output from the foreign matter detection means 53 together with the non-fine foreign matter, and includes a signal indicating the determination result and the fine foreign matter. The image data of the device under test 2 is output to the image display means 7.

  The inspection information input means 6 is composed of, for example, an input device having a keyboard, a mouse, and the like, and information such as setting values and inspection algorithms for specifying the product number and inspection conditions of the DUT 2 (hereinafter referred to as “inspection information”). Is input to the image processing apparatus 5.

  The image display means 7 includes, for example, a liquid crystal display, and displays a determination result of the foreign matter determination means 54 and an X-ray image of the DUT 2 including minute foreign matter.

  Next, the operation of the X-ray foreign object detection device 10 in this embodiment will be described with reference to FIGS. FIG. 3 is a flowchart of the X-ray foreign object detection device 10 in the present embodiment.

  The inspection information input means 6 inputs inspection information to the image processing apparatus 5 (step S1), and the image restoration means 52 of the image processing apparatus 5 performs image processing on the data of the measured object image input from the image input means 51. The restoration process is executed (step S2), and the restored image data is output to the foreign matter detection means 53.

  The foreign matter detection means 53 detects the fine foreign matter mixed in the object to be measured from the restored image together with the non-fine foreign matter (step S3), and outputs the foreign matter image data including the fine foreign matter to the foreign matter determination means 54.

  The foreign matter determining means 54 determines whether or not it is a foreign matter (step S4), and outputs a signal and image data indicating the determination result to the image display means 7.

  The image display means 7 displays the determination result of foreign object detection and the X-ray image of the object to be measured (step S5).

  Next, the image restoration process in step S2 described above will be described in detail with reference to FIGS. FIG. 4 is a flowchart showing image restoration processing in the present embodiment.

  The parameter setting unit 521 of the image restoration unit 52 sets data of parameters N, σ, and γ, and the PSF generation unit 522 generates a function approximate PSF with the set parameter data (step S11).

  The image input unit 51 inputs data of the measurement object image from the X-ray detector 4 (step S12) and outputs the data to the deconvolution unit 523.

  The deconvolution means 523 generates a restored image by performing image restoration processing by deconvolution operation using the measured object image as a convolution image based on a Gaussian function (step S13), and stores the restored image data as an edge storage smoothing means. The first edge preserving smoothing unit 524a of 524 outputs the result.

  The first edge preserving / smoothing unit 524a generates an average value image obtained by weighting the restored image data equally within the square window of the kernel size N ′ with the adaptive moving average filter (step S14). The standard deviation S of the density value in the square window is obtained.

  Further, the first edge preserving / smoothing unit 524a compares the standard deviation S of the density values with the first threshold value A (step S15).

  In step S15, when the first edge preserving smoothing unit 524a determines that the standard deviation S of the density values is equal to or less than the first threshold value A, the first edge preserving smoothing unit 524a outputs the average value image data as it is to the second edge preserving smoothing unit 524b. (Step S16).

  On the other hand, in step S15, if the first edge preserving smoothing unit 524a does not determine that the density value standard deviation S ≦ the first threshold value A (standard deviation S> first threshold value A), The image data P obtained by the correction shown in the equation 2) is output to the second edge preserving / smoothing unit 524b (step S17).

  The second edge preserving / smoothing unit 524b compares the density change value of the image generated by the first edge preserving smoothing process (steps S14 to S17) with the second threshold B (step S14-17). S18).

  In step S18, when the second edge storage smoothing unit 524b determines that the density change value ≦ the second threshold value B, the image data generated by the first edge storage smoothing process is directly used as the first edge storage. The data is output to the smoothing unit 524a.

  On the other hand, in step S18, if the second edge preserving smoothing unit 524b does not determine that the density change value ≦ the second threshold value B (when the density change value> the second threshold value B), the bilateral filter performs the above-described operation. The average value image data weighted by the bi-weighting coefficient W shown in equation (4) is output to the first edge preserving smoothing unit 524a (step S19).

  The first edge preserving / smoothing unit 524a performs the first edge preserving / smoothing process performed in steps S14 to S17 again (step S20), and the second edge preserving / smoothing unit 524b The second edge preserving smoothing process performed in S18 and S19 is performed again (Steps S21 and S22), and the image restoration unit 52 outputs the processed image data to the foreign matter detection unit 53 (Step S23).

  As described above, the edge preserving smoothing unit 524 performs the first edge preserving smoothing process (steps S14 to S19) by the first edge preserving smoothing unit 524a and the second edge preserving smoothing unit 524b, A feature is that the second edge preserving smoothing process (steps S20 to S22) is executed.

  Next, the processed image in the main steps of this embodiment is shown in FIG. FIG. 5 corresponds to the original image of the object to be measured in which minute foreign matter exists in the two circles of FIG.

  FIG. 5A shows a restored image generated by the deconvolution means 523 performing image restoration processing by deconvolution operation in step S13. As shown in FIG. 5A, it can be understood that it is difficult to detect a minute foreign object only by image restoration processing by deconvolution calculation.

  FIG. 5B shows an image after the first edge preserving smoothing process shown in steps S14 to S19. As shown in FIG. 5 (b), the first edge preserving smoothing process characterizes and shows a signal indicating a fine foreign material than that shown in FIG. 5 (a). Since similar signals appear in addition to the above, it can be understood that it is difficult to accurately identify the minute foreign matter.

  FIG. 5C shows an image after the second edge preserving smoothing process shown in steps S20 to S22. As shown in FIG. 5 (c), it can be seen that the second edge preserving smoothing process clearly shows the signal indicating the minute foreign matter at the location where the minute foreign matter is present, so that the minute foreign matter can be easily detected. .

  The above-described operation steps are programmed and stored in a storage medium, and the microcomputer can be caused to function as the image processing device 5 or the X-ray foreign matter detection device 10 by causing the microcomputer to read and execute the program, for example. .

  As described above, according to the image processing apparatus 5 in the present embodiment, the first edge-preserving smoothing that performs the first and second edge-preserving smoothing processes after the deconvolution means 523 that performs the image restoration process, respectively. Edge preserving smoothing means 524 having a converting section 524a and a second edge preserving smoothing section 524b, and the edge preserving smoothing means 524 performs the first and second edge preserving smoothing processes twice each sequentially. Therefore, a higher-definition image can be acquired at a higher speed than the conventional one.

  In addition, according to the X-ray foreign object detection device 10 in the present embodiment, since the image processing device 5 that can acquire a higher-definition image than the conventional one is provided at high speed, scattering blur occurs. Small foreign objects that are buried in Poisson noise and could not be detected by conventional devices can be detected in real time.

(Second Embodiment)
First, the configuration in the second embodiment of the present invention will be described.

  As shown in FIG. 6, the X-ray foreign object detection device 20 in the present embodiment changes a part of the image processing device 5 of the X-ray foreign object detection device 10 (see FIG. 1) in the first embodiment, and performs image processing. The apparatus 8 is used. Therefore, the same reference numerals are given to the same components as those of the X-ray foreign object detection device 10 in the first embodiment, and the description of the configurations will be omitted.

  The image processing apparatus 8 in this embodiment includes an image restoration unit 81, and the image restoration unit 81 includes a first edge storage smoothing unit 811, a deconvolution unit 812, and a second edge storage smoothing unit 813. Yes. The edge preserving / smoothing means 811 and 813 are provided with first edge preserving / smoothing units 811a and 813a and second edge preserving / smoothing units 811b and 813b.

  The first edge preserving smoothing units 811a and 813a and the second edge preserving smoothing units 811b and 813b are respectively the first edge preserving smoothing unit 524a and the second edge preserving smoothing in the first embodiment. The configuration is similar to that of the portion 524b (see FIG. 1).

  Next, image restoration processing by the image restoration means 81 will be described with reference to FIGS. FIG. 7 is a flowchart showing the image restoration processing in the present embodiment. The same steps as those in the first embodiment (see FIG. 4) are denoted by the same reference numerals, and the description of the operation is omitted.

  The image restoration unit 81 performs the first edge storage smoothing process (steps S31 to S33) on the data of the object image input from the image input unit 51 by the first edge storage smoothing unit 811. This process is the same as the process of steps S14 to S19 in the first embodiment.

  The deconvolution means 812 generates a restored image by performing an image restoration process by a deconvolution operation using the image subjected to the first edge preserving smoothing process as a convolution image based on a Gaussian function (step S34). To the edge preserving / smoothing means 813.

  The second edge preserving / smoothing means 813 performs the second edge preserving / smoothing process (steps S35 to S40) on the restored image. This process is the same as the process in steps S20 to S22 in the first embodiment, but the threshold values in steps S36 and S39 are different from those in the first embodiment.

  That is, in the second edge preserving smoothing process, in the third edge preserving smoothing process (steps S35 to S38), in step S36, the first edge preserving smoothing unit 811a performs the standard deviation S of density values. 'And the third threshold value are compared. In step S39, the density change of the image after the third edge preserving smoothing process with respect to the object image is compared with the fourth threshold value. Here, the third threshold value and the fourth threshold value are obtained in advance by including the standard test piece of the minute foreign matter in the DUT 2 as in the first embodiment. Is different from the threshold used for the first edge preserving smoothing process (steps S31 to S33) so that the image is captured most clearly. Note that the threshold setting is not limited to this, and the same threshold may be used for the first and second edge preserving smoothing processes.

  Next, the processed image in the main steps of this embodiment is shown in FIG. FIG. 8 corresponds to the original image of the object to be measured in which minute foreign matter exists in the two circles of FIG.

  FIG. 8A shows an image after the first edge preserving smoothing process shown in steps S31 to S33. As shown in FIG. 8A, it can be understood that the detection of the minute foreign matter is difficult because the signal indicating the minute foreign matter is buried in the Poisson noise only by the first edge preserving smoothing process.

  FIG. 8B shows a restored image generated by performing the image restoration process by the deconvolution unit 812 on the image after the first edge preserving smoothing process in step S34. As shown in FIG. 8B, in the image restoration process based on the deconvolution calculation, a signal indicating a minute foreign matter appears more characteristic than that shown in FIG. 8A. Since a similar signal also appears, it can be understood that it is difficult to accurately identify the minute foreign matter.

  FIG. 8C shows an image after the second edge preserving smoothing process shown in steps S35 to S40. As shown in FIG. 8C, it can be seen that the second edge preserving smoothing process clearly shows the signal indicating the minute foreign matter at the location where the minute foreign matter is present, and the minute foreign matter can be easily detected. .

  As described above, according to the image processing apparatus 8 of the present embodiment, the first edge preserving smoothing unit 811 that performs the first edge preserving smoothing process, and the deconvolution operation after the first edge preserving smoothing process. The deconvolution unit 812 that performs image restoration processing according to the above and the second edge preservation smoothing unit 813 that performs second edge preservation smoothing processing after the image restoration processing are provided. Can acquire high-speed images.

  In addition, according to the X-ray foreign object detection device 20 in the present embodiment, since the image processing device 8 that can obtain a higher-definition image at a higher speed than the conventional one is provided, scattering blur occurs. Small foreign objects that are buried in Poisson noise and could not be detected by conventional devices can be detected in real time.

(Third embodiment)
First, the configuration in the third embodiment of the present invention will be described.

  As shown in FIG. 9, the X-ray foreign object detection device 30 in the present embodiment changes a part of the image processing device 5 of the X-ray foreign object detection device 10 (see FIG. 1) in the first embodiment, and performs image processing. The apparatus 9 is used. Therefore, the same reference numerals are given to the same components as those of the X-ray foreign object detection device 10 in the first embodiment, and the description of the configurations will be omitted.

  The image processing apparatus 9 in this embodiment includes an image restoration unit 91, and the image restoration unit 91 includes an edge preserving smoothing unit 911. The edge preserving / smoothing means 911 includes an edge preserving / smoothing unit 911a.

  The edge preserving / smoothing unit 911a includes a bilateral filter and generates average value image data weighted by the bi-weighting coefficient W shown in the above-described equation (4).

  Next, image restoration processing by the image restoration unit 91 will be described with reference to FIGS. FIG. 10 is a flowchart showing the image restoration process in the present embodiment. The same steps as those in the first embodiment (see FIG. 4) are denoted by the same reference numerals, and the description of the operation is omitted.

  The image restoration unit 91 performs two edge preservation smoothing processes (steps S41 and S42) on the restored image data input from the deconvolution means 523 by the edge preservation smoothing unit 911a. This process is the same as performing the process of step S19 or S22 in the first embodiment twice.

  Next, the processed image in the main steps of this embodiment is shown in FIG. FIG. 11 corresponds to an original image of an object to be measured in which minute foreign objects are present in the two circles of FIG.

  FIG. 11A shows a restored image generated by the deconvolution means 523 performing image restoration processing by deconvolution operation in step S13. As shown in FIG. 11A, it can be understood that it is difficult to detect a minute foreign object only by image restoration processing by deconvolution calculation.

  FIG. 11B shows an image after the first edge preserving smoothing process shown in step S41. As shown in FIG. 11 (b), the first edge preserving smoothing process characterizes and shows a signal indicating a fine foreign material than that shown in FIG. 11 (a). Since similar signals appear in addition to the above, it can be understood that it is difficult to accurately identify the minute foreign matter.

  FIG. 11C shows an image after the second edge preserving smoothing process shown in step S42. As shown in FIG. 11C, it is understood that the signal indicating the minute foreign matter appears clearly at the location where the minute foreign matter is present by the second edge preserving smoothing process, and the minute foreign matter can be easily detected. .

  As described above, according to the image processing apparatus 9 in the present embodiment, the edge preserving smoothing unit 911 having the edge preserving smoothing unit 911a is provided at the subsequent stage of the deconvolution means 523 that performs the image restoration processing, and the edge preserving smoothing is performed. Since the converting means 911 is configured to perform the edge preserving smoothing process twice on the restored image by the bilateral filter, it is possible to obtain a remarkably high-definition image faster than the conventional one.

  Further, according to the X-ray foreign object detection device 30 in the present embodiment, since the image processing device 9 that can acquire a remarkably high-definition image at a higher speed than the conventional one is provided, scattering blur occurs. As a result, it is possible to detect in real time a minute foreign object that is buried in Poisson noise and cannot be detected by a conventional device.

  As described above, the present invention has an effect that a high-definition image can be acquired at a higher speed than the conventional one, and an X-ray is transmitted through a moving object to be measured based on its transmission state. The present invention is useful for an image processing apparatus that detects minute foreign substances contained in an object to be measured, an X-ray foreign object detection apparatus including the same, and an image processing method.

FIG. 1 is a block diagram of a first embodiment of an X-ray foreign object detection device according to the present invention. FIG. 2 is a perspective view of a PSF model in the first embodiment of the X-ray foreign object detection device according to the present invention. FIG. 3 is a flowchart of the first embodiment of the X-ray foreign object detection device according to the present invention. FIG. 4 is a flowchart of image restoration processing in the first embodiment of the X-ray foreign object detection device according to the present invention. FIG. 5 is a processed image in the first embodiment of the X-ray foreign object detection device according to the present invention. FIG. 6 is a block diagram of a second embodiment of the X-ray foreign object detection device according to the present invention. FIG. 7 is a flowchart of image restoration processing in the second embodiment of the X-ray foreign object detection device according to the present invention. FIG. 8 is a processed image in the second embodiment of the X-ray foreign object detection device according to the present invention. FIG. 9 is a block diagram of a third embodiment of the X-ray foreign object detection device according to the present invention. FIG. 10 is a flowchart of image restoration processing in the third embodiment of the X-ray foreign object detection device according to the present invention. FIG. 11 is a processed image in the third embodiment of the X-ray foreign object detection device according to the present invention. FIG. 12A is an original image showing the position of a minute foreign object buried in Poisson noise. FIG. 12B is a restored image by the Wiener filter.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 X-ray source 2 Measured object 3 Conveyance means 4 X-ray detector 5, 8, 9 Image processing apparatus 6 Inspection information input means 7 Image display means 10, 20, 30 X-ray foreign material detection apparatus 51 Image input means (image data) Input means)
52, 81, 91 Image restoration means 53 Foreign matter detection means 54 Foreign matter determination means 521 Parameter setting means 522 PSF generation means (function approximate point spread function generation means)
523, 812 Deconvolution means (restored image generation means)
524, 811, 813, 911 Edge preserving smoothing means 524a, 811a, 813a First edge preserving smoothing unit 524b, 811b, 813b Second edge preserving smoothing unit 911a Edge preserving smoothing unit

Claims (9)

Image data input means (51) for inputting data of a measured object image including edge information indicating an edge of a detection target in an image obtained by photographing the measured object (2);
Function approximate point spread function generating means for generating a function approximate point spread function by approximating a point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard test piece of the detection object (522),
A restored image generating means (523) for making the measured object image a convolution image based on a Gaussian function and performing an image restoration process by a deconvolution operation based on the function approximate point spread function to generate a restored image;
Edge preservation smoothing means (524) for removing noise from the restored image in a state where the edge information is preserved ;
The edge preserving / smoothing means (524) includes first and second edge preserving smoothing units (524a, 524b) for performing first and second edge preserving smoothing processes, respectively.
The first edge preserving smoothing unit (524a), as the first edge preserving smoothing process, performs a simple average of the local densities when the standard deviation of the local density including the neighboring pixels of the target pixel is equal to or less than a first threshold value. Is output to the second edge preserving smoothing unit (524b), and when the standard deviation of the local density exceeds the first threshold, a simple average of the local density is determined in advance. Image data corrected by the correction function to obtain the density of the target pixel is output to the second edge storage smoothing unit (524b),
The second edge preserving / smoothing unit (524b), as the second edge preserving smoothing process, performs the measurement of the image related to the image data output from the first edge preserving / smoothing unit (524a). When the density change value for the image exceeds the second threshold value, the image data subjected to the edge preserving smoothing process by the bilateral filter is output, and when the density change value is not more than the second threshold value, the first edge preserving smoothing unit (524a) outputs the image data. Output the image data without performing the edge preserving smoothing process,
The image processing apparatus according to claim 1, wherein the edge preserving / smoothing means (524) performs the first and second edge preserving smoothing processes twice in sequence . The image processing apparatus according to claim 1, wherein the edge preserving smoothing unit (911) includes an edge preserving smoothing unit (911a) that performs the edge preserving smoothing process a plurality of times by a bilateral filter . Image data input means (51) for inputting data of a measured object image including edge information indicating an edge of a detection target in an image obtained by photographing the measured object (2);
Function approximate point spread function generating means for generating a function approximate point spread function by approximating a point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard test piece of the detection object (522),
Edge preserving smoothing means having first and second edge preserving smoothing units (811a and 811b) for performing first and second edge preserving smoothing processes for removing noise in a state where the edge information is preserved (respectively) 811),
The image related to the image data output by the edge preserving smoothing means (811) is set as a convolution image based on a Gaussian function, and the image is restored by performing image restoration processing by deconvolution operation based on the function approximate point spread function. Restored image generating means (812) for generating
The first edge preserving smoothing unit (811a) performs a simple average of the local densities when the standard deviation of the local density including the neighboring pixels of the target pixel is equal to or less than a first threshold as the first edge preserving smoothing process. Is output to the second edge preserving smoothing unit (811b), and when the standard deviation of the local density exceeds the first threshold, a simple average of the local density is determined in advance. The image data corrected by the correction function to obtain the density of the target pixel is output to the second edge storage smoothing unit (811b),
The second edge preserving / smoothing unit (811b), as the second edge preserving / smoothing process, the device under test for an image related to the image data output by the first edge preserving / smoothing unit (811a). When the density change value for the image exceeds the second threshold, the image data subjected to the edge preserving smoothing process by the bilateral filter is output. When the density change value is equal to or less than the second threshold, the first edge preserving smoothing unit (811a) outputs the image data. Output the image data without performing the edge preserving smoothing process,
The edge preserving / smoothing means (811) performs the first and second edge preserving smoothing processes twice in sequence with the image restoration process in between. apparatus. The bilateral filter is based on a super Gaussian function defined by a parameter based on a distance between a target pixel and a neighboring pixel and a sub-Gaussian function defined by a parameter based on a density difference between the target pixel and the neighboring pixel. The image processing apparatus according to claim 1, wherein the image processing apparatus is a dual weighted moving average filter in which weighting is set . The Gaussian function includes a sub-Gaussian function and a super Gaussian function,
The function approximate point spread function generating means (522) generates one of the sub-Gaussian function and the super Gaussian function by the parameter to generate the function approximate point spread function. The image processing apparatus according to any one of claims 1 to 4. The image processing apparatus according to any one of claims 1 to 5, an X-ray source (1) for irradiating the object to be measured with X-rays, and X transmitted through the object to be measured (2) X-ray detection means (4) for detecting a line and outputting detection information to the image processing apparatus,
The X-ray foreign matter detection device, wherein the image processing device includes foreign matter detection means (53) for detecting foreign matter contained in the object to be measured (2) based on the detection information. An image data input step for inputting data of a measured object image including edge information indicating an edge of a detection target in an image obtained by photographing the measured object;
Function approximate point spread function generation step of generating a function approximate point spread function by approximating the point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard test piece of the detection object When,
An image restoration step for generating a restored image by performing an image restoration process by a deconvolution operation based on the function approximate point spread function based on the measured object image as a convolution image based on a Gaussian function;
An edge preserving smoothing step for removing noise from the restored image in a state where the edge information is preserved,
The edge preserving smoothing step includes first and second edge preserving smoothing steps for performing first and second edge preserving smoothing processes, respectively.
In the first edge preserving smoothing step, as the first edge preserving smoothing process, when the standard deviation of the local density including neighboring pixels of the target pixel is equal to or less than a first threshold, the simple average of the local density is used as the target Image data having pixel density is output, and when the standard deviation of the local density exceeds the first threshold, the simple average of the local density is corrected by a predetermined correction function to obtain the density of the target pixel. Output
In the second edge preserving smoothing step, as the second edge preserving smoothing process, a density change value with respect to the measured object image of an image related to the image data output in the first edge preserving smoothing step is obtained. When the second threshold value is exceeded, the image data subjected to the edge preserving smoothing process by the bilateral filter is output, and when it is equal to or less than the second threshold value, the image data from the first edge preserving smoothing step is output to the edge preserving smoothing process. Output without doing
In the edge preserving smoothing step, each of the first and second edge preserving smoothing processes is sequentially performed twice, respectively. The image processing method according to claim 7, wherein the edge preserving smoothing step includes an edge preserving smoothing step in which edge preserving smoothing is performed a plurality of times by a bilateral filter . An image data input step for inputting data of a measured object image including edge information indicating an edge of a detection target in an image obtained by photographing the measured object;
Function approximate point spread function generation step of generating a function approximate point spread function by approximating the point spread function with a parametric function based on a parameter determined based on an edge image acquired in advance using a standard test piece of the detection object When,
An edge preserving smoothing step having first and second edge preserving smoothing steps for performing first and second edge preserving smoothing processes for removing noise in a state where the edge information is preserved;
An image related to the image data output in the edge preserving smoothing step is a convolution image based on a Gaussian function, and a restored image is generated by performing an image restoration process by a deconvolution operation based on the function approximate point spread function. Image restoration step,
In the first edge preserving smoothing step, as the first edge preserving smoothing process, when the standard deviation of the local density including neighboring pixels of the target pixel is equal to or less than a first threshold, the simple average of the local density is used as the target Image data having pixel density is output, and when the standard deviation of the local density exceeds the first threshold, the simple average of the local density is corrected by a predetermined correction function to obtain the density of the target pixel. Output
In the second edge preserving smoothing step, as the second edge preserving smoothing process, a density change value with respect to the measured object image of an image related to the image data output in the first edge preserving smoothing step is obtained. When the second threshold value is exceeded, the image data subjected to the edge preserving smoothing process by the bilateral filter is output, and when it is equal to or less than the second threshold value, the image data from the first edge preserving smoothing step is output to the edge preserving smoothing process. Output without doing
In the edge preserving / smoothing step, the first and second edge preserving / smoothing processes are sequentially performed twice each with the image restoration step in between .