Image processing method and apparatus for reducing noise contained in images acquired by radiography

Abstract

an image processing device for performing noise reduction processing on an image obtained by radiation imaging using a recursive method, the image processing device including: an image input unit for receiving an input of an image for each frame obtained by radiation imaging; a motion map generating unit for generating a motion map including motion detection information for each pixel of a difference map obtained by subtracting a current frame image from a previous frame image; a cumulative motion map generating unit for generating a cumulative motion map based on the generated motion map and a motion map accumulated up to a previous frame; a statistical value map generating unit for generating a statistical value map formed by assigning a statistical value indicating a motion probability of each pixel to each pixel of the cumulative motion map; and an output image generating unit for generating an output image of the current frame by mixing the current frame image and an output image of the previous frame based on the cumulative motion map.

Description

[Technical Field]

[0001] The present invention relates to an image processing method and apparatus for reducing noise contained in an image acquired by radiography. [Background technology]

[0002] Various medical diagnoses and treatments are performed based on images captured by radiographic imaging devices that use radiation such as X-rays. Medical images captured using X-rays generally contain not only anatomical information of the subject, i.e., the human body, but also noise generated by the imaging environment and equipment performance. This noise degrades the image and reduces the ability to interpret the patient's anatomy. The degree of image degradation is determined by various factors, including image degradation due to noise from the sensor itself and noise generated by the X-ray dose.

[0003] However, prolonged exposure to radiation generated during X-ray imaging can cause side effects due to radiation exposure. For this reason, it is necessary to take X-ray images using low-dose X-rays to minimize radiation exposure. However, when taking images using low-dose X-rays, the photon density of the incident X-rays decreases, resulting in a significant increase in the concentration of photon mottle, which can lead to a decrease in image quality.

[0004] For this reason, various methods have been introduced to effectively remove noise from acquired images while using low-dose X-rays. For example, Korean Patent No. 10-1432864 introduces a noise reduction technology using recursive filtering, in which a noise component image is obtained by thresholding and reducing a difference map obtained by subtracting the image of the current frame from the image of the previous frame, and then this is added to the image of the current frame to generate an output image of the current frame, and the noise-reduced output image is stored in memory.

[0005] However, while recursive filtering-based noise reduction techniques are highly effective at removing line noise and detecting stationary objects without motion, they suffer from the drawback that their noise reduction performance deteriorates with increasing object motion. In particular, incorrectly identifying object motion can result in motion blur, degrading image quality. Therefore, in recursive filtering, it is crucial to set a threshold value for distinguishing object motion from noise in a difference map between the current and previous frames. In Korean Patent Registration No. 10-1432864, object motion is determined to exist if the absolute value of each pixel in the difference map exceeds a preset threshold value that is greater than the standard deviation of the noise; otherwise, object motion is determined to be absent. However, due to the characteristics of X-ray images, noise intensity and deviation vary significantly depending on the dose conditions, object density, and the degree of object motion during imaging, resulting in a wide range of threshold values and making it difficult to set an appropriate threshold value. Furthermore, the presence of object motion reduces the effectiveness of recursive filtering, reducing noise reduction. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Korean Patent Registration No. 10-1432864 [Patent Document 2] Patent No. 6744440 Summary of the Invention [Problem to be solved by the invention]

[0007] The problem to be solved by the present invention is to provide a method for effectively reducing noise by using a cumulative motion map in which images that have undergone noise reduction processing based on the motion comparison results in a difference map are accumulated in chronological order. [Means for solving the problem]

[0008] An image processing device according to an embodiment of the present invention performs noise reduction processing on an image obtained by radiography using a recursive method, and includes: an image input unit for receiving frame-by-frame images obtained by radiography, a motion map generation unit for generating a motion map including motion detection information for each pixel of a difference map obtained by subtracting a current frame image from a previous frame image, a cumulative motion map generation unit for generating a cumulative motion map based on the generated motion map and a motion map accumulated up to a previous frame, a statistical map generation unit for generating a statistical map by assigning a statistical value indicating a motion probability of each pixel to each pixel of the cumulative motion map, and an output image generation unit for generating an output image for the current frame by blending the current frame image and an output image for the previous frame based on the cumulative motion map.

[0009] The motion map generation unit may be configured to utilize the statistical value map to generate the motion map.

[0010] The motion map generation unit may include a difference map generation unit that generates the difference map based on a difference between the current frame image and the previous frame image, and a threshold processing unit that generates the motion map including the motion sensing information of each pixel by thresholding pixel values of each pixel of the difference map.

[0011] The thresholding unit may be configured to threshold the difference map through an adaptive threshold whose magnitude depends on the magnitude of the statistics of the statistical value map.

[0012] The thresholding unit may be configured to set a lower threshold for each pixel for the thresholding process as the statistical value of the corresponding pixel in the statistical value map indicates a higher probability of motion.

[0013] The statistical value map generation unit may be configured to calculate the statistical value in units of a pixel mask including a plurality of pixels including each pixel of the cumulative motion map, and assign the calculated statistical value as a statistical value for each pixel to generate the statistical value map.

[0014] The statistic may be the entropy, variance or standard deviation of pixel values of a plurality of pixels of the pixel mask.

[0015] The output image generation unit may be configured to determine a blending ratio of the current frame image and the previous frame output image based on a cumulative motion map updated using the statistical value map.

[0016] The cumulative motion map generation unit may be configured to threshold the statistical value map and use the thresholded statistical value map to selectively increase or decrease pixel values of pixels in the cumulative motion map to update the cumulative motion map.

[0017] The difference map may be obtained using one or more images of the current frame image and a previous frame image.

[0018] The previous frame image may be the output image of the previous frame.

[0019] Meanwhile, the cumulative motion map generation unit may be configured to update the cumulative motion map using the statistical value map.

[0020] According to an embodiment of the present invention, the output image generation unit may be configured to determine a ratio for blending the current frame image and the previous frame output image based on a cumulative motion map updated using the statistical value map.

[0021] An image processing device according to an embodiment of the present invention, which performs noise reduction processing on an image obtained by radiography using a recursive method, includes the steps of receiving an input of a frame-by-frame image obtained by radiography, generating a motion map including motion detection information for each pixel of a difference map obtained by subtracting a current frame image from a previous frame image, generating a cumulative motion map based on the generated motion map and a motion map accumulated up to the previous frame, generating a statistical value map by assigning, to each pixel of the cumulative motion map, a statistical value indicating the motion probability of a plurality of pixels including the pixel, and generating an output image of the current frame by blending the current frame image and an output image of the previous frame based on the cumulative motion map.

[0022] In the generating of the motion map, the motion map may be generated using the statistical value map.

[0023] In the motion map generating step, the motion map including the motion sensing information of each pixel may be generated by thresholding the pixel value of each pixel of the difference map, and the thresholding may be configured to be performed by an adaptive threshold whose magnitude varies depending on the magnitude of the statistical value of the statistical value map.

[0024] In the motion map generation step, the motion map including the motion sensing information of each pixel may be generated by thresholding the pixel value of each pixel of the difference map, and the threshold value for each pixel for the thresholding may be set lower as the motion probability indicated by the statistical value of the corresponding pixel in the statistical value map increases.

[0025] In the generating of the statistical map, the statistical map may be generated by calculating the statistical values in units of a pixel mask including a plurality of pixels including each pixel of the cumulative motion map and allocating the calculated statistical values as the statistical values of the respective pixels.

[0026] The statistic may be the entropy, variance or standard deviation of pixel values of a plurality of pixels of the pixel mask.

[0027] The step of generating the output image may include updating the cumulative motion map using the statistical value map, and determining a blending ratio of the current frame image and the previous frame output image based on the updated cumulative motion map.

[0028] In updating the cumulative motion map, the statistical value map may be thresholded, and the cumulative motion map may be updated by selectively increasing or decreasing pixel values of pixels in the cumulative motion map using the thresholded statistical value map.

[0029] The difference map may be obtained by subtracting the current frame image from the output image of the previous frame.

[0030] Meanwhile, an image processing method according to another embodiment of the present invention may further include updating the cumulative motion map using the statistical value map.

[0031] According to an embodiment of the present invention, in generating the output image, a mixing ratio of the current frame image and the previous frame output image may be determined based on the updated cumulative motion map. [Effects of the Invention]

[0032] According to the present invention, it is possible to generate an image with reduced noise through a cumulative motion map. [Brief explanation of the drawings]

[0033] [Figure 1] 1 is a diagram illustrating an example of a C-arm type radiation imaging apparatus to which an image processing apparatus according to an embodiment of the present invention is applied; [Figure 2]1 is a block diagram of a video processing device according to an embodiment of the present invention; [Figure 3] 2 is a flowchart of an image processing method according to an embodiment of the present invention. [Figure 4] 10 is a graph showing a comparison of the pixel value distribution of the difference map of the original signal and the pixel value distribution of the difference map obtained after the Anscombe transform for the same input video (previous frame and current frame) of a stationary subject. [Figure 5] 2 is a diagram illustrating an algorithm of an image processing method according to an embodiment of the present invention; [Figure 6] An example of a motion map output when the hand phantom moves from right to left in the previous frame video and the current frame video is shown. [Figure 7] 10 is a diagram illustrating a method for generating a cumulative motion map using a motion map. [Figure 8] 10 is a diagram illustrating generation of a motion map and a cumulative motion map according to an input image as frames progress; [Figure 9] An example of a motion map and output image obtained when a thin lead rod passes over a stationary pelvis phantom is shown. [Figure 10] 9 shows examples of motion maps when the threshold for motion detection is increased and decreased under the same conditions as in FIG. [Figure 11] 10 shows an example of an entropy map generated by calculating entropy in a pixel masking operation using a cumulative motion map. [Figure 12] 10 is a diagram illustrating a method for generating a motion map according to another embodiment of the present invention. [Figure 13] 10 is a diagram illustrating an algorithm of an image processing method according to another embodiment of the present invention; [Figure 14] 14 is a diagram illustrating a process of updating a cumulative motion map using a thresholded entropy map in the image processing method of FIG. 13; DETAILED DESCRIPTION OF THE INVENTION

[0034] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] The present invention will now be described in detail with reference to the accompanying drawings, so that those skilled in the art can easily understand the present invention. However, the present invention may be embodied in many different forms and is not limited to the described embodiments.

[0035] An example of a C-arm type radiographic imaging apparatus to which an image processing device having a noise reduction function according to an embodiment of the present invention is applied is shown in Figure 1. The image processing device according to an embodiment of the present invention may be applied to a part of the image capturing apparatus as shown in Figure 1, or may be configured as a separate device from the image capturing apparatus to perform the image processing function.

[0036] An image capturing device to which an image processing device according to an embodiment of the present invention is applied may be configured as an image capturing device capable of acquiring video, for example, in the form of a C-arm as shown in FIG. 1, and may be configured to capture an area of interest of a subject S, which is the object of imaging, using radiation such as X-rays.

[0037] 1, the radiographic apparatus may include a radiation emitting unit 110 that emits radiation, for example, X-rays, and an image acquiring unit 120 that detects radiation that has passed through a subject S and acquires image data, and the radiation emitting unit 110 and the image acquiring unit 120 may be supported on both ends of a C-arm 210. For example, the radiographic apparatus may be used as a mobile C-arm X-ray imaging device, an interventional X-ray device, an interventional angiography C-arm X-ray device, etc.

[0038] The support structure 200 supports the radiation emitting unit 110 and the image acquiring unit 120 and is configured to change the spatial and rotational positions of the radiation emitting unit 110 and the image acquiring unit 120 in order to change the imaging position and angle of the irradiated object S. For example, the support structure 200 may include a support body 240, a lift column 230 connected to the support body 240 so as to be movable in a vertical direction D1, and a forward / backward moving arm 220 connected to the lift column 230 so as to be movable in a vertical direction D2 relative to the lift column 230.

[0039] The C-arm 210 is fastened to the forward / backward movement arm 220 so as to be rotatable relative to the forward / backward movement arm 220 in at least one rotational direction, and the radiation irradiation unit 110 and the image acquisition unit 120 are fastened to both ends of the C-arm 210. In this case, the C-arm 210 can move up and down together with the forward / backward movement arm 220 and move forward and backward in the horizontal direction, and can also be fastened to the forward / backward movement arm 220 so as to be rotatable relative to the forward / backward movement arm 220 in at least one rotational direction, for example, an orbital rotational direction R1 and an axial rotational direction R2 centered on a direction parallel to the horizontal movement direction of the forward / backward movement arm 220. Although not shown in the drawings, the support structure 200 may include actuators such as motors for vertical movement of the lift column 230, horizontal movement of the forward / backward movement arm 220, and rotation of the C-arm 210. The elements for supporting and driving the C-arm 210, which is a support member that supports the radiation irradiator 110 and the image acquirer 120, i.e., the forward / backward arm 220, the lift column 230, and the actuators provided thereon, can be said to be driving elements for driving the C-arm 210, and a combination of these can be said to be a driving unit for driving the C-arm 210. In addition, the forward / backward arm 220 can be configured to enable panning rotation of the C-arm 210 through lateral rotation. The shape of the support member is not limited to a C-shape, and in other embodiments of the present invention, arms shaped like U-shapes, G-shapes, etc. can be used as the support member instead of C-shapes.

[0040] The display unit 140 is configured to display at least one of real-time position information, image data, reference position information, and radiation output information. The display unit 140 may be any device capable of displaying information and images, such as a printer, a CRT display, an LCD display, a PDP display, an OLED display, an FED display, an LED display, a DLP display, a PFD display, a 3D display, a transparent display, etc. The display unit 140 may also be implemented in a form capable of displaying and inputting information, such as a touch screen that can receive input from a user.

[0041] The image processing device 10 according to an embodiment of the present invention may be implemented in the form of one or more information processing devices, such as computers, capable of processing information and performing calculations. For example, the computer may include a control unit such as a CPU, a storage unit such as a ROM (read only memory) or a RAM (random access memory), and a graphics control unit such as a GPU (graphics processing unit). The computer may also include a communication unit such as a network card, and input / output units such as a keyboard, a display, or a touch screen. These computer components may be connected via a bus, as is well known, and may be operated and controlled by the execution of a program stored in the storage unit.

[0042] The image processing device 10, which may be implemented in the form of a computer capable of information processing, may be installed in the radiation imaging device shown in FIG. 1 and configured to perform an image processing function. In this case, the image processing device 10 may be configured to receive and process images captured as part of the radiation imaging device, and display the processed images on an image display means of the radiation imaging device.

[0043] 2, an image processing device 10 according to an embodiment of the present invention includes a plurality of functional components for noise reduction processing, namely, an image input unit 11, a difference unit 14, a threshold processing unit 15, a cumulative motion map generation unit 16, and an output image generation unit 17. A motion map is generated by the sequential operations of the difference unit 14 and the threshold processing unit 15, and in this sense, the difference unit 14 and the threshold processing unit 15 can be considered as a motion map generation unit.

[0044] 3, an image processing method according to an embodiment of the present invention that can be performed by the image processing device 10 may include an image acquisition step (S31), a difference map generation step (S34), a motion map generation step (S35), a cumulative motion map generation step (S36), a statistical value map (e.g., an entropy map) generation step (S38), and an image output step (S37). The image processing method generates an output image with reduced noise through a recursive method. In this regard, this method is also called infinite recursive depth.

[0045] The image input unit 11 receives an input image from the outside. For example, the image input unit 11 may receive frames of a moving image acquired by the image acquiring unit 120 in time sequence. An image consisting of a plurality of pixels of an area including an object may be acquired as a radiation moving image sequence by the image acquiring unit 120, which is a radiation detection panel, in time sequence, and the acquired radiation moving image sequence may be input to the image input unit 11 in time sequence.

[0046] The input image may be one frame of a video consisting of multiple frames. The input image may also be a two-dimensional image having multiple pixels consisting of multiple rows (n rows) and multiple columns (m columns), and may include noise along with the captured object. For example, the noise included in the input image may include line noise and random noise in a predetermined direction, such as the horizontal and / or vertical direction.

[0047] A motion map is generated by generating a difference map by the difference unit 14 and thresholding the difference map by the threshold processing unit 15. As a result, a motion map can be generated from the noise-processed current frame image and the previous frame image, e.g., the output image of the previous frame, as shown in FIG. 5 . The motion map is obtained from data obtained by subtracting the previous frame image from the current frame image and includes information on the presence or absence of motion for each pixel. For example, pixels with motion can be set to have a value of “0” and pixels without motion can be set to have a value of “1.” That is, all pixels in the motion map have a pixel value of 0 or 1, and a pixel with a value of 0 indicates the presence of motion relative to the previous frame image, while a pixel with a value of 1 indicates the absence of motion relative to the previous frame image. The presence or absence of motion can be determined based on threshold processing. In an embodiment of the present invention, a statistical value map, e.g., an entropy map, is used in the threshold processing for motion determination.

[0048] The difference unit 14 generates a difference map by subtracting the current frame image from the previous frame image, e.g., the output image of the previous frame. The difference can be performed after noise processing, e.g., noise reduction and noise stabilization, is performed on the current frame image and the output image of the previous frame. By using the output image of the previous frame to generate the difference map for the current frame, a so-called recursive filtering structure of noise reduction processing can be performed. The difference unit 14 can generate the difference map by calculating the pixel value difference of each pixel at the same position for all pixels between the current frame image and the previous frame image. That is, the difference map includes the same number of pixels as the frame image, and the value of each pixel in the difference map corresponds to the difference value of that pixel between the current frame image and the previous frame image. It should be understood that the previous frame image is a concept that includes not only the image of the frame immediately before the current frame, but also the image of the frame before that. The difference map can be obtained by subtracting the current frame image, e.g., the input image, from the image of one or more previous frames, e.g., the output image. For example, the difference map can be generated by subtracting one of the output images of the previous frames from the input image of the current frame, or by subtracting a mixed image obtained by mixing the output images of multiple previous frames from the input image of the current frame.

[0049] The difference map may include information about the motion of the object and information about residual noise. Additionally, the generated difference map may be stabilized by using a mean filter or a median filter.

[0050] The threshold processing unit 15 determines a threshold for detecting motion in the difference map, performs threshold processing based on the determined threshold, and detects the presence or absence of motion at each pixel, thereby generating a motion map. That is, the threshold processing unit 15 determines the presence or absence of motion at each pixel in the difference map through threshold processing, and generates a motion map including information indicating the presence or absence of motion at each pixel based on the threshold. If the threshold for detecting motion is set too low, the motion detection sensitivity increases and the noise reduction level decreases. Conversely, if the threshold is set too high, the motion detection sensitivity decreases and motion blur (a phenomenon of blurring) may occur. Because X-ray images are acquired under different dose conditions and subject characteristics, it is difficult to predict the pixel values of the acquired images, and therefore it is necessary to set an appropriate threshold based on the pixel values.

[0051] The threshold for motion detection can be determined by various methods, such as statistical methods, experimental methods, and arithmetic methods. Figure 4 compares the pixel value distribution of the difference map of the original signal and the pixel value distribution of the difference map obtained after the Anscombe transform for the same input video (previous frame and current frame) of a stationary object. Figure 4(a) is a graph showing the pixel value distribution of the difference map obtained from the original signal, and Figure 4(b) is a graph showing the pixel value distribution of the difference map obtained after the Anscombe transform of the original signal. In the graph of Figure 4, the horizontal axis (x-axis) represents pixel value (intensity), and the vertical axis (y-axis) represents the pixel value difference between pixels at the same position in the previous and current frame images. Because the object is stationary, the data distribution in the y-axis direction for any x-value can be considered noise, and if the data distribution in the y-axis direction is equal to or greater than the noise value, it can be assumed that the pixel has motion.

[0052] For example, in the distribution of the difference map of the original signal, if the pixel value is 500, an appropriate threshold for motion determination is approximately 100, and if the pixel value is 2200, an appropriate threshold is approximately 170. As such, the appropriate threshold for motion determination varies greatly depending on the pixel value, making it difficult to determine the threshold. In contrast, in the distribution of the difference map of the Anscombe transformed signal, as shown in FIG. 4(b), if the pixel value is 500, an appropriate threshold is approximately 5, and if the pixel value is 2200, an appropriate threshold is approximately 4. As such, when the difference map is generated by differentiating the Anscombe transformed signal, the difference in appropriate thresholds depending on the pixel value becomes very small, making it very easy to determine the threshold. In this case, for example, the threshold may be set to a value that falls within the top 70 to 80% range of the distribution of pixel value differences for a particular pixel value.

[0053] Even if the difference data is calculated from the Anscombe transformed signal, there will still be some deviation when applying the same threshold to all pixel values. Taking this into consideration, adaptive thresholding can be used to more fairly correct noise deviations for all pixel values. The reason why the difference value changes nonlinearly depending on the pixel value is due to the mathematical characteristics of the Anscombe transform.

[0054] The threshold processing unit 15 generates a motion map by performing threshold processing for motion detection on the difference map generated by the difference unit 14. The threshold processing unit 15 generates a motion map from the difference map by applying an adaptive threshold using a statistical map, e.g., an entropy map, generated by a statistical map (e.g., entropy map) generation unit 18, which will be described later. A method of determining whether motion is detected using the adaptive threshold and generating a motion map based on the determination will be described later.

[0055] The threshold processing unit 15 determines that each pixel in the difference map has motion if its pixel value is equal to or greater than a predetermined threshold and assigns a judgment value indicating motion, such as "0," to the pixel. Conversely, if its pixel value is less than the threshold, the threshold processing unit 15 determines that the pixel has no motion and assigns a judgment value indicating no motion, such as "1," to the pixel. The threshold processing unit 15 performs this motion judgment using a threshold for all pixels in the difference map and assigns a judgment value to the pixel according to the result to generate a motion map. In this manner, the threshold processing unit 15 generates a motion map using the difference map and a threshold reference curve indicating an adaptive threshold. To this end, the motion map has the same pixels as the input frame image and a value indicating the presence or absence of motion for each pixel.

[0056] Meanwhile, in another embodiment of the present invention, thresholding may be performed not on a pixel-by-pixel basis but on a mask-by-mask basis, which includes multiple adjacent pixels. For example, a mask including 3*3 or 9 pixels surrounding a central pixel may be set, and the average or median value of the pixels in the mask may be compared with a threshold to determine whether or not there is motion at the central pixel. In this manner, motion may be determined on a mask-by-mask basis while moving from one mask to another, and a motion map may be generated based on the determination. By performing thresholding on a mask-by-mask basis, noise may be reduced if it is present in a specific pixel.

[0057] As described above, the threshold processing unit 15 determines whether or not each pixel has motion and assigns different values to pixels determined to have motion and pixels determined to have no motion, so that all pixels in the difference map have one of two predetermined values, for example, 0 or 1. The map obtained in this way contains information on whether or not each pixel has motion, and in this sense, this map can be called a motion map.

[0058] Figure 6 shows an example of a motion map output when a hand phantom moves from right to left in a previous frame image and a current frame image. Figure 6(a) shows a previous frame image containing a hand phantom, and Figure 6(b) shows a current frame image in which the hand phantom has moved from right to left. Figure 6(c) shows an output motion map showing the motion detection results for each pixel of the difference data.

[0059] Referring to (c) of Figure 6, the white areas indicate pixels where motion was detected, and it can be seen that areas in the hand phantom where there is actual motion that are highly contrasted with the previous frame (the periphery of the bone area) are detected as motion, while areas where there is little contrast with the previous frame (skin area, tool area) are not detected as motion. As a result, when generating the final output image in the subsequent process, if the areas with motion are updated around the current frame and the remaining areas are averaged, noise can be reduced, edges can be strengthened, sharpness can be improved, and motion blur can be prevented.

[0060] The cumulative motion map generation unit 16 accumulates motion maps in a separate memory in chronological order based on the motion detection result and adds them together to generate a cumulative motion map, which is used to determine an appropriate blending ratio of the previous frame image and the current frame image to generate a final output image.

[0061] FIG. 7 illustrates a method for generating a cumulative motion map using a motion map. The cumulative motion map generation unit 16 generates and updates a cumulative motion map by accumulating the motion maps obtained through motion detection in a separate memory in chronological order for each frame. The image on the left side of FIG. 7 shows a cumulative motion map, and the image on the right side shows a cumulative motion map obtained by adding together the motion maps. The cumulative motion map for the current frame can be generated by adding together the values of the same pixel in the cumulative motion map for the previous frame and the motion map for the current frame. In other words, the cumulative motion map for a particular frame is an image obtained by adding together the pixel values of the same pixel in all motion maps obtained up to that frame. For example, referring to FIG. 7, if the current frame is the fourth frame, the sum of the values of the same pixel in the motion maps obtained from the previous four frames becomes the value of the same pixel in the cumulative motion map for the current frame.

[0062] If a pixel determined to have motion in a motion map is represented brightly as "0" and a pixel determined to have no motion is represented darkly as "1," each pixel in the cumulative motion map has a value corresponding to the sum of the motion detection values (0 or 1) of that pixel, and has a different value, i.e., different brightness, depending on the number of motion detections. For example, assuming a cumulative motion map consisting of 10 frames, the cumulative motion map is obtained by adding up the 10 motion maps, and each pixel in the cumulative motion map has a value between 0 and 10. Here, if the same pixel in all motion maps has a value of 0, that pixel in the cumulative motion map has a value of 0, and if the same pixel in all motion maps has a value of 10, that pixel in the cumulative motion map has a value of 10. Thus, as shown in FIG. 8, each pixel in the cumulative motion map has a pixel value, i.e., brightness, corresponding to the number of motion detections of the same pixel in the motion maps up to the frame.

[0063] In this sense, the brighter the pixel value in the generated cumulative motion map, the higher the motion probability, and the darker the pixel value, the lower the motion probability. For example, if motion is detected at the same pixel position while frames are progressing in chronological order, the motion probability will be high, and if motion is not detected, the motion probability will be low.

[0064] The value of each pixel in the cumulative motion map indicates the degree of motion, or motion probability value, of that pixel. In other words, if the motion map for each frame assigns a judgment value of "0" if there is motion and "1" if there is no motion, the smaller the pixel value in the cumulative motion map, the greater the motion probability of that pixel.

[0065] Meanwhile, in another embodiment of the present invention, instead of accumulating pixel values of the motion map when updating the cumulative motion map, the cumulative motion map may be updated by selectively increasing or decreasing pixel values of pixels in the cumulative motion map using a thresholded statistical value map.

[0066] 8 shows an example of a process for generating a motion map and a cumulative motion map according to input images as frames progress. A motion map is generated based on the difference between the frame image of each frame and the previous frame image, and the cumulative motion map for the current frame is generated by combining the generated motion map with the cumulative motion map for the previous frame.

[0067] Meanwhile, according to an embodiment of the present invention, the entropy map generation unit 18 generates a statistical value map for adaptive thresholding used to generate a motion map. The statistical value map is generated from the cumulative motion map and is formed to have the same number of pixels as the cumulative motion map. Each pixel in the statistical value map has a statistical value indicating the probability of motion occurring at that pixel in the cumulative motion map as its pixel value. For example, a mask including multiple pixels including the pixel in the cumulative motion map can be set, and a statistical value indicating the probability of motion occurring in the mask can be set as the statistical value of the pixel. For example, the statistical value indicating the probability of motion occurring at a particular pixel can be the entropy value or standard deviation of the pixel values of multiple pixels surrounding the particular pixel. Hereinafter, an example will be described in which entropy is used as the statistical value, and in this sense, the statistical value map will be referred to as an entropy map. The entropy map is generated from the cumulative motion map using the concept of entropy, a statistical approach. The generated entropy map is stored in a separate buffer memory and can be used to generate a motion map for the next frame. The motion map of the current frame can be generated using the entropy map generated in the previous frame, and if necessary, the motion map of the current frame can be generated again using the entropy map generated in the current frame. In this case, the motion map of the current frame can be generated using not only the entropy map of the immediately preceding frame but also one or more entropy maps of the previous frame.

[0068] Since the motion map is generated based on a difference map obtained by subtracting the current frame image from the previous frame image, the motion of the object may not be accurately detected in some cases. For example, if the object is thin, it may be difficult to detect motion if the difference value, i.e., the pixel value of each pixel of the difference map, is not sufficiently greater than the noise level. For example, when a thin, sharp object such as a needle passes through a thick part of a patient, it may be difficult to detect the motion of the needle. To solve this problem, in an embodiment of the present invention, a motion map is generated by applying a threshold value having a different value to each pixel depending on conditions through adaptive thresholding.

[0069] FIG. 9 shows a motion map and output image obtained when a thin lead rod passes over a stationary pelvis phantom. This image was obtained without adaptive thresholding according to an embodiment of the present invention. (a) of FIG. 9 shows the motion map, and (b) of FIG. 9 shows the output image. The arrow in (b) of FIG. 9 indicates the direction of motion. Referring to FIG. 9, it can be seen that in the thin parts of the phantom where the rod passes, the difference between before and after the rod passes is large enough to detect the motion as a pixel. However, when the rod passes through a thick bone region, the difference is not large enough to detect the motion as a pixel. In this case, the pixel is determined to be noise and averaged out when generating the output image, resulting in a force image in which the rod appears to be cut off.

[0070] Figure 10 shows motion maps obtained under the same conditions as Figure 9. Figure 10(a) shows the motion map when the threshold for motion detection is increased, and Figure 10(b) shows the motion map when the threshold is lowered. As shown in Figure 10(b), when the threshold for motion detection is lowered so that the stick does not break, the noise is regarded as moving pixels, resulting in a side effect of being emphasized. As such, there is a trade-off depending on the threshold when detecting the motion of a thin object.

[0071] To address these issues, an embodiment of the present invention utilizes the concept of entropy, a statistical approach. Entropy is a numerical representation of the degree of certainty or amount of information contained in a probability distribution. When the probability of a specific value occurring in a probability distribution is high and the probability of the remaining values is low, the entropy decreases. Conversely, when the probabilities of various values occurring are similar, the entropy increases. Entropy can be considered one of the characteristic values that indicate the distribution characteristics of a probability distribution. When the probability or probability density is concentrated around a specific value, the entropy is said to be low. Conversely, when the probability or probability density is uniformly spread across various values, the entropy is said to be high. In other words, the higher the entropy, the greater the heterogeneity and the greater the amount of information. Meanwhile, the lower the entropy, the greater the homogeneity and the less the amount of information.

[0072] As is well known, in statistics, entropy (H) means the average amount of information, and can be defined as follows for a discrete random variable X:

[0073]

number

[0074] The entropy of the motion probability values may be calculated for each pixel or for each group of pixels in the cumulative motion map, and an entropy map may be generated by assigning the calculated entropy to each pixel or a group of pixels. For example, a masking area may be selected for all pixels in the cumulative motion map, the entropy values of the motion probability values of the pixels belonging to the masking area may be calculated, and the calculated entropy value may be assigned as the pixel value of the pixel to generate the entropy map. A sufficiently large area including the pixel to be calculated may be taken as a mask, and the entropy may be calculated using a histogram for the motion probability values within the mask. In this case, the histogram may be calculated by using the motion probability values as frequencies and the number of occurrences of each motion probability value as frequencies, and the entropy may be calculated based on this using Equation 1. For example, if the cumulative motion map is a 1500*1500 image, a 300*300 mask including the center pixel to be calculated may be set, and the entropy within the mask may be calculated and set as the entropy of the center pixel. Figure 11 shows an example of an entropy map generated by calculating entropy through a pixel masking operation in a cumulative motion map. Figure 11(a) shows an example of a cumulative motion map, with the masks shown as dotted lines. Entropy is calculated for each mask, and of the three masks shown in Figure 11(a), the rightmost mask contains the most bright pixels, i.e., pixels with a high motion probability. Therefore, it can be predicted that the pixels belonging to the rightmost mask have the highest entropy. Similarly, in Figure 11(a), the leftmost mask contains the most dark pixels, i.e., pixels with a low motion probability. Therefore, it can be predicted that the pixels belonging to the leftmost mask have the lowest entropy. Figure 11(b) shows an example of an entropy map generated by calculating the entropy for the masks set for all pixels in the cumulative motion map of (a).As a result, each pixel in the entropy map has the entropy value of the motion probability values of pixels belonging to the mask that includes the pixel as its pixel value. In this case, a pixel or area with a large entropy can be considered to indicate a large probability of motion, and a pixel or area with a small entropy can be considered to indicate a small probability of motion. In this sense, entropy can be considered a measure of the degree of congestion of motion probability values.

[0075] In an embodiment of the present invention, the generated entropy map is used to generate a motion map, and a threshold for motion detection is adjusted using the entropy of each pixel included in the entropy map. The generated motion map is stored in a separate buffer memory, and adaptive thresholding is performed based on this entropy map, i.e., the entropy map generated in the previous frame, when generating a motion map for the next frame. The threshold for each pixel used in thresholding performed by the thresholding unit can be set to be inversely proportional to the magnitude of the entropy of the corresponding pixel in the entropy map. For example, when generating a motion map, pixels or regions with high entropy have a high probability of having previously moved, so the motion detection threshold can be lowered to detect motion more sensitively. Conversely, pixels or regions with low entropy have a low probability of having previously moved, so the motion detection threshold can be increased to detect motion more conservatively. This adaptive thresholding can improve the trade-off problem depending on the threshold value mentioned above. Meanwhile, if variance or standard deviation is used as a statistical value instead of entropy, pixels or regions with small variance or standard deviation can be considered to have a high probability of having previously moved.

[0076] By generating a motion map through adaptive threshold processing based on entropy in this way, the accuracy of motion detection for individual frames is increased. Even if accurate motion detection is not performed for each frame, the amount of information accumulates as the frames progress, and the area gradually expands in the direction of lowering the threshold of the surrounding pixels of the moving object. As a result, the accuracy of motion detection for thin objects is significantly improved as the frames progress.

[0077] 12 shows a method for generating a motion map according to another embodiment of the present invention. Although the motion map is generated through threshold processing according to the noise level of the image in the above embodiment, noise components may remain in the motion map. Therefore, in this embodiment, an additional noise reduction process is performed.

[0078] In this embodiment, after separating the subject from the background, remaining noise is removed by excluding the background area from the motion map. A background image from which the subject has been removed is generated by experimentally applying a threshold to the current frame image or through AI learning, and a noise-free motion map is generated by adding or multiplying the motion map generated according to the above embodiment with the generated background image. This removes noise components remaining in the background, allowing for the generation of a motion map containing only information about the subject's motion. This method has the advantage that, when the motion detection threshold is lowered, the noise level of the entire image increases slightly, while the motion of the subject can be detected more sensitively.

[0079] Meanwhile, in another embodiment of the present invention, when setting a threshold for motion detection, two motion maps may be generated by separating the background and subject areas and applying different thresholds to each area, and then the two generated motion maps may be merged to generate a final motion map.

[0080] The output image generation unit 17 generates an output image by mixing the current frame image and the previous frame image. At this time, the output image generation unit 17 generates a final output image by mixing the current frame image and the previous frame output image at an appropriate mixing ratio based on the cumulative motion map. The reflection ratio of the current frame image may be set to be relatively higher as the value of each pixel in the cumulative motion map indicates motion. That is, the mixing ratio may be determined such that the weight of the current frame increases as the motion probability determined by the pixel values included in the cumulative motion map increases. For example, the reflection ratio may be determined such that the reflection ratio of the current frame image increases linearly as the pixel value of the cumulative motion map indicates motion on a pixel-by-pixel basis.

[0081] As a specific example, for pixels with a lot of motion, a predetermined weight (α1), e.g., 0.8, is assigned to the pixel in the current frame image, and a weight (1-α1), e.g., 0.2, is assigned to the pixel in the accumulated output image up to the previous frame. This blending process is performed for all pixels to blend the current frame image and the previous frame output image. On the other hand, for pixels with little motion, a predetermined weight (α2), e.g., 0.2, is assigned to the pixel in the current frame image, and a weight (1-α2), e.g., 0.8, is assigned to the pixel in the accumulated previous frame image to blend the current frame image and the previous frame output image. A higher current frame weight for pixels with a lot of motion results in an image without motion blur. As a result, the final output image is updated to reflect more of the current frame's value based on motion probability for pixels with a lot of motion, and updated to reflect more of the previous frame's value for pixels with little motion. This improves noise reduction performance as frames accumulate over time.

[0082] After the final noise-reduced image is output, the result is stored in memory and used as the previous frame image when generating the output image for the next frame, thereby implementing a recursive method.

[0083] 13 illustrates the process of an image processing method according to another embodiment of the present invention. As shown in FIG. 13, the entropy map generated by the cumulative motion map is used when generating an output image by blending the current frame image and the previous frame image. As in the previous embodiment, the blending ratio of the current frame image and the previous frame image is determined by the cumulative motion map, and pixels with high entropy values indicate a high probability that motion has occurred in the previous frame.

[0084] Through thresholding of the entropy map, pixels where motion is determined to have occurred can be determined. The threshold for thresholding can be determined experimentally in advance, or sensitivity can be set by viewing the image quality displayed in real time through a user interface. For example, when generating an entropy map, once the mask size for entropy calculation is determined, the maximum entropy value can be determined, and the lower 20% of the maximum entropy value can be set as the threshold.

[0085] In this embodiment, the entropy map is thus thresholded and the thresholded entropy map is used to update the cumulative motion map, as illustrated in FIG.

[0086] Although the embodiments of the present invention have been described above, the scope of the present invention is not limited thereto, and includes all changes and modifications that can be easily modified from the embodiments of the present invention by a person having ordinary knowledge in the technical field to which the present invention belongs and are recognized as equivalents. [Industrial Applicability]

[0087] INDUSTRIAL APPLICABILITY The present invention relates to a method for processing images obtained by radiography, and is applicable to image processing technology, and therefore has industrial applicability.

Claims

1. An image processing device that performs noise reduction processing on an image obtained by radiography using a recursive method, an image input unit for receiving an input of frame-by-frame images obtained by the radiation imaging; a motion map generating unit for generating a motion map including motion sensing information for each pixel of a difference map obtained by subtracting a current frame image from a previous frame image; a cumulative motion map generating unit for generating a cumulative motion map based on the generated motion map and a motion map accumulated up to a previous frame; a statistical map generating unit for generating a statistical map by assigning a statistical value indicative of a motion probability of each pixel to each pixel of the cumulative motion map; an output image generating unit that generates the output image of the current frame by mixing the current frame image and the output image of the previous frame based on the cumulative motion map; the cumulative motion map generation unit is configured to threshold the statistical map and to update the cumulative motion map by selectively increasing or decreasing pixel values of pixels in the cumulative motion map using the thresholded statistical map; The image processing device is configured such that the output image generation unit determines a blending ratio of the current frame image and the output image of the previous frame based on an updated cumulative motion map using the statistical value map.

2. The motion map generation unit a difference map generating unit for generating the difference map based on a difference between the current frame image and the previous frame image; The image processing device of claim 1 , further comprising: a thresholding unit configured to generate the motion map including the motion sensing information of each pixel by thresholding a pixel value of each pixel of the difference map.

3. The image processing device of claim 2 , wherein the thresholding unit is configured to threshold the difference map through an adaptive threshold whose magnitude varies depending on the magnitude of the statistics of the statistical value map.

4. The image processing device of claim 2 , wherein the threshold processing unit is configured to set a lower threshold value for each pixel for the threshold processing as the statistical value of the corresponding pixel in the statistical value map indicates a higher motion probability.

5. The image processing device of claim 1 , wherein the statistical value map generation unit is configured to calculate the statistical value in units of a pixel mask including a plurality of pixels including each pixel of the cumulative motion map, and assign the calculated statistical value as a statistical value for each pixel to generate the statistical value map.

6. The image processing device of claim 5 , wherein the statistical value is an entropy, a variance, or a standard deviation of pixel values of a plurality of pixels in the pixel mask.

7. The image processing device of claim 1 , wherein the difference map is obtained using at least one of the current frame image and a previous frame image.

8. The image processing device according to claim 1 , wherein the previous frame image is an output image of the previous frame.

9. An image processing method for performing noise reduction processing on an image obtained by radiography using a recursive method, comprising: receiving an input of frame-by-frame images obtained by the radiation imaging; generating a motion map including motion detection information for each pixel of a difference map obtained by subtracting a current frame image from a previous frame image; generating a cumulative motion map based on the generated motion map and a motion map accumulated up to a previous frame; a statistical map generating step of generating a statistical map by assigning a statistical value indicating a motion probability of a plurality of pixels including each pixel to each pixel of the cumulative motion map; generating an output image of the current frame by blending the current frame image and the output image of the previous frame based on the cumulative motion map; The step of generating the output image comprises: updating the cumulative motion map using the statistical value map; determining a blending ratio of the current frame image and the output image of the previous frame based on the updated cumulative motion map.

10. 10. The image processing method of claim 9, wherein, in the motion map generating step, the motion map including the motion sensing information of each pixel of the difference map is generated by thresholding pixel values of the pixels of the difference map, and the thresholding is performed using an adaptive threshold whose magnitude varies depending on the magnitude of the statistical value of the statistical value map.

11. 10. The image processing method of claim 9, wherein, in the motion map generating step, the motion map including the motion sensing information of each pixel is generated by thresholding pixel values of each pixel of the difference map, and the threshold value for each pixel used in the thresholding is set lower as the motion probability indicated by the statistical value of the corresponding pixel in the statistical value map increases.

12. 10. The image processing method of claim 9, wherein in the generating the statistical map, the statistical map is generated by calculating the statistical values in units of a pixel mask including a plurality of pixels including each pixel of the cumulative motion map and allocating the calculated statistical values as the statistical values of the respective pixels.

13. The image processing method of claim 12 , wherein the statistical value is an entropy, a variance, or a standard deviation of pixel values of a plurality of pixels of the pixel mask.

14. 10. The image processing method of claim 9, wherein, in updating the cumulative motion map, the statistical value map is thresholded, and the cumulative motion map is updated by selectively increasing or decreasing pixel values of pixels in the cumulative motion map using the thresholded statistical value map.

15. The image processing method of claim 9 , wherein the difference map is obtained using at least one of the current frame image and a previous frame image.

16. The image processing method according to claim 9 , wherein the difference map is obtained by subtracting the current frame image from the output image of a previous frame.

Images