Description:BACKGROUND
Technical Field
[0001] The embodiments herein relate to the field of X-ray image processing and machine learning, more specifically, to a system and method for dual-energy X-ray image denoising with a quality score from a machine learning (ML) model determining the stopping criterion.
Description of the Related Art
[0002] Material discrimination from X-ray images requires high-quality imaging to capture detailed images, advanced noise reduction to enhance quality without losing critical details, and material-specific calibration for accurate identification. The existing systems for material discrimination rely on noise reduction techniques that are often resource-intensive and introduce latency, while sometimes losing valuable signal details or adding artifacts. The existing systems for material discrimination frequently lack customization, scalability, and robustness, making it difficult for them to adapt to specific applications, larger datasets, or X-ray system-specific noise. Integrating the existing systems into existing infrastructure is complex and costly, and has high maintenance requirements.
[0003] Further, the existing systems for material discrimination encounter errors due to the noise in the captured intensities and conventional denoising methods fail to denoise the high-energy image leading to poor estimates of material properties. The existing system also make the image appear less detailed and less distinct, which is problematic in applications where fine details are essential for accurate analysis or interpretation, such as material discrimination or object detection.
[0004] Accordingly, there remains a need to address the technical problem of X-ray image denoising for material discrimination applications. Given the fact that there is no metric to measure the quality of an X-ray image, we design a machine learning model which takes as input an X-ray image and gives a quality score that is indicative of the noise in the image.
SUMMARY
[0005] In view of the foregoing, embodiments herein provide a method for X-ray image denoising using machine learning (ML) models to estimate the quality (noise level) of the image which in turn is used to minimize an interference /increase performance in material discrimination. The method includes (1) obtaining a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) of an object from an X-ray detector, where the object is made of one or more materials, (2) implementing a first set of variance stabilization (VST) and an anisotropic diffusion technique on the LEI to generate a guidance LEI by reducing noise and preserving edges of the LEI, (3) implementing a second set of VST and the anisotropic diffusion technique on the guidance LEI based on a Laplace function of the guidance LEI to generate a filtered LEI and one or more edge parameters (E), (4) generating a filtered HEI using the filtered LEI and one or more edge parameters (E) by implementing a guided filtering technique on the filtered LEI based on an inverse variance stabilization of the filtered LEI by removing noise, (5) assigning a first quality score to the filtered LEI and a second quality score to the filtered HEI by estimating the noise based on a standard deviation of signal-dependent noise, the signal-dependent noise is Poisson in nature or follows Poisson distribution, and (6) generating a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise if the first quality score and the second quality score are greater than a predetermined threshold score, thereby improving the performance of material discrimination.
[0006] The method is of advantage that the method enables the preservation of edge information, by determining where diffusion occurs and where diffusion is restricted. The method enables the creation of an optimal guide image for filtering both low-energy and high-energy X-ray images. Additionally, the no-reference image quality measure designed functions as a mechanism for incorporating a stopping condition. The method achieves a higher level of denoising compared to individual denoising of the low-energy and high-energy X-ray images.
[0007] Further, the method is of advantage that the method allows for reducing signal-dependent noise in X-ray images, which is generated during the photon-counting process and adversely affects material discrimination. By managing the signal-dependent noise, the method increases the clarity and accuracy of the X-ray images. Furthermore, the method improves reliability of material discrimination, enabling more precise and dependable results in applications such as medical imaging, security screening, and industrial inspections.
[0008] In some embodiments, the method includes assigning successive first-quality scores to the filtered LEI and successive second-quality scores to the filtered HEI to reduce the signal-dependent Poisson noise until the optimized performance of material discrimination is obtained.
[0009] In some embodiments, the one or more materials are aluminum, nylon, Teflon, or, stainless steel.
[0010] In some embodiments, the method includes obtaining the LEI and the HEI by fixing the position of the object on a belt for estimating ground truth to calculate the first quality score and the second quality score.
[0011] In some embodiments, the variance stabilization (VST) method is applied to the LEI to extract edge parameters (the coefficient of the anisotropic diffusion equation which decides the amount of diffusion) of the LEI for the anisotropic diffusion. The edge parameters are coefficients in the diffusion equation that indicate the amount of diffusion.
[0012] In some embodiments, the filtered LEI acts as a guidance image to filter the HEI.
[0013] In some embodiments, the method includes training the machine learning model that is an SVM regressor by providing the parameters of a generalized Gaussian as features of the LEI and the HEI as training data with ground truth of the dual-energy X-ray image derived by averaging the captured LEI and HEI to implement the first quality score and the second quality score.
[0014] In some embodiments, the method includes enhancing material discrimination performance by reducing the standard deviation of Zeff (effective atomic number), where the effective atomic number (Zeff) measures the net positive charge felt by an electron and assists in material discrimination by highlighting differences in atomic structure and electron distribution.
[0015] In some embodiments, the quality score acts as a stopping criterion for iterative denoising of the dual-energy X-ray image and describes the extent of denoising of the dual-energy X-ray image achieved.
[0016] In one aspect, a system for dual-energy X-ray image processing using machine learning models to estimate the quality (noise level) of the image which in turn is used to minimize an interference /increase performance in material discrimination by reducing Poisson noise in an X-ray image is provided. The system includes a quality improvement and assessment server that receives a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) of an object from an X-ray detector, where the object is made of one or more materials, where the quality improvement and assessment server includes a memory that stores a set of instructions and a processor that is configured to execute the set of instructions, where the processor is configured to (1) implement a first set of variance stabilization (VST) and an anisotropic diffusion technique on the LEI to generate a guidance LEI by reducing noise and preserving edges of the LEI, (2) implement a second set of VST and the anisotropic diffusion technique on the guidance LEI based on a Laplace function of the guidance LEI to generate a filtered LEI and one or more edge parameters (E), (3) generate a filtered HEI using the filtered LEI and the one or more edge parameters (E) by implementing a guided filtering technique on the filtered LEI based on an inverse variance stabilization of the filtered LEI by removing noise, (4) assign a first quality score to the filtered LEI and a second quality score to the filtered HEI by estimating the noise based on a standard deviation of signal-dependent noise, where the signal-dependent noise is Poisson in nature or follows Poisson distribution, and (5) generate a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise if the first quality score and the second quality score are greater than a predetermined threshold score, thereby improving the performance of material discrimination.
[0017] The system is of advantage that the system enables the preservation of edge information, by determining of where diffusion occurs and where diffusion is restricted. The system enables the creation of an optimal guide image for filtering both low-energy and high-energy X-ray images. Additionally, the no-reference image quality measure designed functions as a mechanism for incorporating a stopping condition. The system achieves a higher level of denoising compared to individual denoising of the low-energy and high-energy X-ray images.
[0018] Further, the system is of advantage that the method allows for reducing signal-dependent noise in X-ray images, which is generated during the photon-counting process and adversely affects material discrimination. By managing the signal-dependent noise, the system increases the clarity and accuracy of the X-ray images. Furthermore, the system improves reliability of material discrimination, enabling more precise and dependable results in applications such as medical imaging, security screening, and industrial inspections.
[0019] These and other aspects of the embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating preferred embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the spirit thereof, and the embodiments herein include all such modifications.
BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The embodiments herein will be better understood from the following detailed description with reference to the drawings, in which.
[0021] FIG. 1 is a block diagram of a system for dual-energy X-ray image processing for reducing Poisson noise in an X-ray image to estimate the performance using ML models according to some embodiments herein;
[0022] FIG. 2 is a block diagram of a quality improvement and assessment server of FIG. 1 according to some embodiments herein;
[0023] FIG. 3 is a block diagram of a comparison of the denoising method of a low-energy X-ray image of materials according to some embodiments herein;
[0024] FIG. 4 is a flow diagram of a comparison of the denoising method of a high-energy X-ray image of materials according to some embodiments herein;
[0025] FIG. 5 is a flow diagram of a method for dual-energy X-ray image processing for reducing Poisson noise in an X-ray image to estimate the performance using ML models; and
[0026] FIG. 6 is a schematic diagram of a computer architecture of the subset deriving server or one or more entity devices in accordance with embodiments herein.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

[0027] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
[0028] There remains a need for a system and method for dual-energy X-ray image processing for reducing Poisson noise in an X-ray image to estimate the performance using ML models.
[0029] Referring now to the drawings, and more particularly to FIGS. 1 to 6, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments.
[0030] The term “Variance Stabilizing Transformation (VST)” refers to a mathematical transformation applied to data to stabilize the variance across different levels or ranges. The Poisson variance being same as its mean VST is needed to convert the samples into constant variance so that the subsequent algortihms like the anisotropic diffusion and guided filter are applicable.
[0031] The term “signal-dependent noise” refers to noise whose characteristics, such as variance and mean, are dependent on the magnitude or level of the underlying signal, which means that the amount or nature of the noise changes with the signal strength. Signal-dependent noise is commonly encountered in low light imaging and X-ray imaging.
[0032] The term “image processing” refers to the manipulation and analysis of digital images using various algorithms and techniques.
[0033] The term “Poisson noise” refers to a statistical noise that follows a Poisson distribution. The Poisson noise is commonly encountered in processes where events occur independently and at a constant average rate, particularly in photon counting and other quantum mechanical processes.
[0034] The term “anisotropic diffusion” is a technique used primarily in image processing for edge-preserving smoothing. Unlike isotropic diffusion, which diffuses equally in all directions, anisotropic diffusion diffuses preferentially in certain directions based on the local image structure. Anisotropic Diffusion enables preservation of important features like edges while reducing noise.
[0035] FIG. 1 is a block diagram of a system 100 for dual-energy X-ray image processing for reducing Poisson noise in an X-ray image and to estimate the performance using ML models according to some embodiments herein. The system 100 includes an X-ray detector 102, a network 104, a quality improvement and assessment server 106, and a machine learning model 108. The X-ray detector 102 captures a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) from objects by fixing the position of the scanned object on the belt. The objects may be a laptop bag, a luggage bag, or a handbag. The objects may be made of a variety of organic and inorganic materials, aluminum material, nylon material, Teflon material, or, stainless steel material to list a few. The X-ray detector 102 is communicatively connected to the quality improvement and assessment server 106 through the network 104. In some embodiments, the network 104 is a wired network. In some embodiments, the network 104 is a wireless network. In some embodiments, the network 104 is a combination of the wired network and the wireless network. In some embodiments, network 104 is the Internet.
[0036] The quality improvement and assessment server 106 includes a machine learning model 108. The quality improvement and assessment server 106 improves the quality using anisotropic and guided filtering and assesses the quality of dual-energy X-ray images affected by signal-dependent noise using the machine learning model 108. The X-ray detector 102 is fixed in the x-ray machine.The signal-dependent noise is due to the photon counting process and follows the Poisson distribution. The dual-energy X-ray image includes a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI). The high-energy image has much more noise than the low-energy X-ray image. The image quality improvement and assessment server 106 utilizes a Variance Stabilizing Transformation (VST) and an Anisotropic Diffusion technique followed by guided filtering to process the dual-energy X-ray image to generate a denoised dual-energy by estimating the noise level of the X-ray image. The reduced noise level minimizes interference and improves performance in material discrimination by reducing Poisson noise in an X-ray image. The material discrimination performance is improved by reducing the standard deviation (SD) of Zeff (effective atomic number). The Zeff measures the net positive charge felt by an electron and assists in material discrimination by highlighting differences in atomic structure and electron distribution. The below Table:1 indicates the standard deviation of the predicted Zeff before and after denoising.
Material SD of Zeff before denoising SD of Zeff after denoising
Nylon 0.72 0.12
Teflon 0.37 0.08
Aluminium 0.8 0.3
Table: 1
[0037] The machine learning model 108 calculates a feature vector from the input image by estimating the correlations from a noramalised image and learns to assign a quality factor using an SVM model.
[0038] The system 100 is of advantage that the system 100 enables the extraction of edge information, enabling the determination of where diffusion occurs and where diffusion is restricted. The system 100 enables the creation of an optimal guide image for filtering both low-energy and high-energy X-ray images. Additionally, the no-reference image quality measure designed functions as a mechanism for incorporating a stopping condition. The system 100 achieves a higher level of denoising compared to individual denoising of the low-energy and high-energy X-ray images.
[0039] Further, the system 100 is of advantage that the method allows for reducing signal-dependent noise in X-ray images, which is generated during the photon-counting process and adversely affects material discrimination. By managing the signal-dependent noise, the system 100 increases the clarity and accuracy of the X-ray images. Furthermore, the system 100 improves reliability of material discrimination, enabling more precise and dependable results in applications such as medical imaging, security screening, and industrial inspections.
[0040] FIG. 2 is a block diagram of a quality improvement and assessment server of FIG. 1 according to some embodiments herein. The quality improvement and assessment server 106 includes an image obtaining module 202, a guidance LEI generating module 204, a filtered LEI and edge parameters generating module 206, a filtered HEI and edge parameters generating module 208, a quality score assigning module 210, and a threshold validating module 212.
[0041] The image obtaining module 202 receives the dual-energy X-ray image from the X-ray detector 102. The dual-energy X-ray image may include a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI). The guidance LEI generating module 204 receives the LEI from the image obtaining module 202. The guidance LEI generating module 204 applies a first set of variance stabilization (VST) and anisotropic diffusion techniques on the LEI to generate a guidance LEI. The VST and anisotropic diffusion techniques work by reducing noise and preserving edges of the LEI, thereby enhancing the quality of the image for subsequent processing tasks. The first set of variance stabilization (VST) and anisotropic diffusion techniques is used to make the noise variance uniform across different signal levels, resulting in a variance-stabilized LEI. The inverse variance stabilised LEI after diffusion, with reduced noise and preserve structural details, serves as a high-quality guide image for the guided filter that significantly improves the performance of subsequent image analysis and interpretation.
[0042] The filtered LEI and edge parameters generating module 206 applies a second set of Variance Stabilization (VST) and anisotropic diffusion techniques to the guidance LEI, using a Laplace function of the guidance LEI to generate a filtered LEI and a set of edge parameters.
[0043] The filtered HEI and edge parameters generating module 208 generates a filtered High-Energy Image (HEI) involves using the filtered Low-Energy Image (LEI) and a set of edge parameters (E) using a guided filtering technique. The process of generating filtered HEI and edge parameters starts by creating a filtered LEI through a second set of Variance Stabilization (VST) and anisotropic diffusion techniques applied to the guidance LEI, using a Laplace function to enhance edge preservation and noise reduction. The edge parameters are extracted during the process of capturing important structural features of the HEI. A guided filtering technique is then applied to the filtered LEI, utilizing the edge parameters to ensure significant features are preserved while noise is reduced. The inverse variance stabilization involves reversing the effect of variance stabilisation. The result is a filtered HEI with effectively removed noise and maintained structural details, suitable for enhanced analysis and processing. The edge parameters are coefficients in the diffusion equation that describe the amount of diffusion. The filtered LEI acts as a guidance image to filter the HEI.
[0044] The quality score assigning module 210 assigns a first quality score to the filtered Low-Energy Image (LEI) and a second quality score to the filtered High-Energy Image (HEI) involves estimating noise levels based on the standard deviation of signal-dependent noise, which follows a Poisson distribution. The machine learning model 108 is trained by an SVM regressor and involves using the features - depending on the statistical properties of the image - of the Low-Energy Image (LEI) and High-Energy Image (HEI) as training data. The ground truth for this training is derived by averaging the captured LEI and HEI to obtain the dual-energy X-ray image. The ground truth is used to train the SVM regressor. The quality scores indicate the effectiveness of noise reduction and the preservation of important image details, providing a measure of the overall quality of the filtered images. The first quality score and the second quality score are calculated by estimating ground truth to calculate. Assigning the successive first quality scores to the filtered Low-Energy Image (LEI) and the successive second quality scores to the filtered High-Energy Image (HEI) involves evaluating and reducing signal-dependent Poisson noise iteratively until optimized performance in material discrimination is achieved.
[0045] The threshold validating module 212 generates a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise involves validating that both the first and second quality scores exceed a predetermined threshold. If the quality scores are higher than the predetermined threshold, the dual-energy X-ray image is considered sufficiently denoised, which enhances the performance of material discrimination. If the quality scores do not exceed the predetermined threshold, the dual-energy X-ray image is filtered again. The quality score acts as a stopping criterion for iterative denoising of the dual-energy X-ray images and describes the extent of denoising of the dual-energy X-ray images achieved.
[0046] FIG. 3 is a block diagram of a comparison of the denoising method of a low-energy X-ray image of materials according to some embodiments herein. The comparison of the denoising method of a Low-Energy X-ray Image between the applicant system with BM3D following Variance Stabilization (VST). The scores for each image are displayed the corresponding image, higher score indicates better performance. It may be noted that the proposed solution performs at par with BM3D for low energy and much better than BM3D for high energy images. Aso, the computation time of the applicant system is smaller by a large factor compared to BM3D. The first column shows the original image along with a zoomed-in section highlighting fine edges. The second column presents the results of BM3D following VST, while the third column shows the results of the applicant system. The applicant system achieves a score comparable to BM3D, but unlike BM3D, it preserves the fine edges effectively. In addition, the time required by the applicant system is much smaller compared with the time needed for BM3D.
[0047] FIG. 4 is a flow diagram of a comparison of the denoising method of a high-energy X-ray image of materials according to some embodiments herein. The comparison of the denoising method of a high-energy X-ray Image between the applicant system with BM3D following Variance Stabilization (VST). The scores for each image are displayed above the corresponding image. The first column shows the original image along with a zoomed-in section highlighting fine edges. The second column presents the results of BM3D following VST, while the third column shows the results of the applicant system. The applicant system achieves a score comparable to BM3D, but unlike BM3D, it preserves the fine edges effectively and has much lower computation time as evident from the Table 3.
[0048] The below Table:2 presents a comparison of the image quality index in terms of score between VST+BM3D and the applicant system:
Index Score for low-energy X-ray image Score for high-energy X-ray image
Original VST+BM3D Applicant Original VST+BM3D Applicant
1 77.23 80.95 80.9 49.81 71.7 76.27
2 49.58 52.59 54.64 56.15 64.98 61.7
3 48.54 52.33 54.31 51.94 59.87 58.1
4 54.32 69.44 74.86 78.38 89.42 90.43
5 43.29 48.53 49.16 36.89 64.23 68.38
6 81.51 91.68 91.03 82.18 90.59 91.15
7 70.33 80.44 84.00 47.39 73.31 77.35
Table: 2
The applicant system outperforms BM3D in most cases. Additionally, the applicant system offers the advantage of denoising the buffers in under 40 milliseconds, whereas BM3D takes several seconds to achieve similar results. The computation time comparison is given in Table 3-
Index
Size VST+BM3D (Time in seconds) Applicant (Time in seconds)
1 528x570 4.6092 0.1423
2 432x360 2.1769 0.0770
3 416x310 1.8210 0.0675
4 304x300 1.4086 0.0375
5 448x540 3.8432 0.1058
6 128x150 0.2865 0.0101
7 456x242 1.6548 0.0485
Table: 3
The comparison tables (2 and 3) above demonstrate that the applicant provides better quality denoised output in a shorter amount of time.

[0049] FIG. 5 is a flow diagram of a method for dual-energy X-ray image denoising with a quality factor decided by a machine learning model to estimate the quality (noise level) of the image which in turn is used to minimize an interference /increase performance in material discrimination by reducing Poisson noise in an X-ray image. a. The method. At step 502, the method includes obtaining a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) of an object from an X-ray detector, wherein the object is made of a plurality of materials. At step 504, the method includes implementing a first set of variance stabilization (VST) and an anisotropic diffusion technique on the LEI to generate a guidance LEI by reducing noise and preserving edges of the LEI. At step 506, the method includes implementing a second set of VST and the anisotropic diffusion technique on the guidance LEI based on a Laplace function of the guidance LEI to generate a filtered LEI and a plurality of edge parameters (E). At step 508, the method includes generating a filtered HEI using the filtered LEI and the plurality of edge parameters (E) by implementing a guided filtering technique on the filtered LEI based on an inverse variance stabilization of the filtered LEI by removing noise. At step 510, the method includes assigning a first quality score to the filtered LEI and a second quality score to the filtered HEI by estimating the noise based on a standard deviation of signal-dependent noise, wherein the signal-dependent noise is Poisson in nature or follows Poisson distribution. At step 512, the method includes generating a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise if the first quality score and the second quality score are greater than a predetermined threshold score, thereby improving the performance of material discrimination.
[0050] The method is of advantage that the method enables the extraction of edge information, enabling the determination of where diffusion occurs and where diffusion is restricted. The method enables the creation of an optimal guide image for filtering both low-energy and high-energy X-ray images. Additionally, the no-reference image quality measure designed functions as a mechanism for incorporating a stopping condition. The method achieves a higher level of denoising compared to individual denoising of the low-energy and high-energy X-ray images.
[0051] Further, the method is of advantage that the method allows for reducing signal-dependent noise in X-ray images, which is generated due to the photon-counting process and adversely affects material discrimination. By managing the signal-dependent noise, the method increases the clarity and accuracy of the X-ray images. Furthermore, the method improves reliability of material discrimination, enabling more precise and dependable results in applications such as medical imaging, security screening, and industrial inspections.
[0052] In embodiments herein may include a computer program product configured to include a pre-configured set of instructions, which when performed, can result in actions as stated in conjunction with the methods described above. In an example, the pre-configured set of instructions can be stored on a tangible non-transitory computer readable medium or a program storage device. In an example, the tangible non-transitory computer readable medium can be configured to include the set of instructions, which when performed by a device, can cause the device to perform acts similar to the ones described here. Embodiments herein may also include tangible and/or non-transitory computer-readable storage media for carrying or having computer executable instructions or data structures stored thereon.
[0053] Generally, program modules utilized herein include routines, programs, components, data structures, objects, and the functions inherent in the design of special-purpose processors, etc. that perform particular tasks or implement particular abstract data types. Computer executable instructions, associated data structures, and program modules represent examples of the program code means for executing steps of the methods disclosed herein. The particular sequence of such executable instructions or associated data structures represents examples of corresponding acts for implementing the functions described in such steps.
[0054] The embodiments herein can include both hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc.
[0055] A data processing system suitable for storing and/or executing program code will include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
[0056] Input/output (I/O) devices (including but not limited to keyboards, displays, pointing devices, etc.) can be coupled to the system either directly or through intervening I/O controllers. Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.
[0057] A representative hardware environment for practicing the embodiments herein is depicted in FIG. 6, with reference to FIGS. 1 through 5. This schematic drawing illustrates a hardware configuration of a server or a computer system or a computing device in accordance with the embodiments herein. The system includes at least one processing device CPU 10 that may be interconnected via system bus 14 to various devices such as a random-access memory (RAM) 15, read-only memory (ROM) 17, and an input/output (I/O) adapter 17. The I/O adapter 17 can connect to peripheral devices, such as disk units 12 and program storage devices 13 that are readable by the system. The system can read the inventive instructions on the program storage devices 13 and follow these instructions to execute the methodology of the embodiments herein. The system further includes a user interface adapter 20 that connects a keyboard 18, mouse 19, speaker 25, microphone 23, and other user interface devices such as a touch screen device (not shown) to the bus 14 to gather user input. Additionally, a communication adapter 21 connects the bus 14 to a data processing network 42, and a display adapter 22 connects the bus 14 to a display device 24, which provides a graphical user interface (GUI) 30 of the output data in accordance with the embodiments herein, or which may be embodied as an output device such as a monitor, printer, or transmitter, for example.
[0058] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and/or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the appended claims.
, C , C , Claims:I/We Claim:
1. A method for dual-energy X-ray image denoising with quality factor determined using a machine learning model (noise level) which in turn is used to minimize noise which in turn increases performance in material discrimination, the method comprises:
obtaining a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) of an object from an X-ray detector (102), wherein the object is made of a plurality of materials;
implementing a first set of variance stabilization (VST) and an anisotropic diffusion technique on the LEI to generate a guidance LEI by reducing noise and preserving edges of the LEI;
implementing a second set of VST and the anisotropic diffusion technique on the guidance LEI based on a Laplace function of the guidance LEI to generate a filtered LEI and a plurality of edge parameters (E);
generating a filtered HEI using the filtered LEI and the plurality of edge parameters (E) by implementing a guided filtering technique on the filtered LEI based on an inverse variance stabilization of the filtered LEI by removing noise;
assigning a first quality score to the filtered LEI and a second quality score to the filtered HEI by estimating the noise based on a standard deviation of signal-dependent noise, wherein the signal-dependent noise is Poisson in nature or follows Poisson distribution; and
generating a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise if the first quality score and the second quality score are greater than a predetermined threshold score, thereby improving the performance of material discrimination.

2. The method as claimed in claim 1, wherein the method comprises assigning successive first quality scores to the filtered LEI and successive second quality scores to the filtered HEI to reduce the signal-dependent Poisson noise until the optimized performance of material discrimination is obtained.


3. The method as claimed in claim 1, wherein the plurality of materials are at least one of an aluminum, a nylon, a Teflon, or, a stainless steel.


4. The method as claimed in claim 1, wherein the method comprises obtaining the LEI and the HEI by fixing the position of the object on a belt for estimating ground truth to calculate the first quality score and the second quality score.


5. The method as claimed in claim 1, wherein the variance stabilization (VST) method is applied to the LEI to extract edge parameters (the coefficient of the anisotropic diffusion equation which decides the amount of diffusion) of the LEI for the anisotropic diffusion, wherein the edge parameters are coefficients in the diffusion equation that indicate the amount of diffusion.


6. The method as claimed in claim 1, wherein the filtered LEI acts as a guidance image to filter the HEI.


7. The method as claimed in claim 1, wherein the method comprises training the machine learning model that is an SVM regressor by providing the parameters of a generalized Gaussian as features of the LEI and the HEI as training data with ground truth of the dual-energy X-ray image derived by averaging the captured LEI and HEI to implement the first quality score and the second quality score.


8. The method as claimed in claim 1, wherein the method comprises enhancing material discrimination performance by reducing the standard deviation of Zeff (effective atomic number), wherein the effective atomic number (Zeff) measures the net positive charge felt by an electron and assists in material discrimination by highlighting differences in atomic structure and electron distribution.


9. The method as claimed in claim 1, wherein the quality score acts as a stopping criterion for iterative denoising of the dual-energy X-ray image and describes the extent of denoising of the dual-energy X-ray image achieved.


10. A system for dual-energy X-ray image processing using machine learning models to estimate the quality (noise level) of the image which in turn is used to minimize an interference /increase performance in material discrimination by reducing Poisson noise in an X-ray image, wherein the system comprises,
a quality improvement and assessment server (106) that receives a low-energy X-ray image (LEI) and a high-energy X-ray image (HEI) of an object from an X-ray detector (102), wherein the object is made of a plurality of materials, wherein the quality improvement and assessment server (106) comprises, a memory that includes a set of instructions and a processor that executes the set of instructions and is configured to:
implement a first set of variance stabilization (VST) and an anisotropic diffusion technique on the LEI to generate a guidance LEI by reducing noise and preserving edges of the LEI;
implement a second set of VST and the anisotropic diffusion technique on the guidance LEI based on a Laplace function of the guidance LEI to generate a filtered LEI and a plurality of edge parameters (E);
generate a filtered HEI using the filtered LEI and the plurality of edge parameters (E) by implementing a guided filtering technique on the filtered LEI based on an inverse variance stabilization of the filtered LEI by removing noise;
assign a first quality score to the filtered LEI and a second quality score to the filtered HEI by estimating the noise based on a standard deviation of signal-dependent noise, wherein the signal-dependent noise is Poisson in nature or follows Poisson distribution; and
generate a denoised dual-energy X-ray image with reduced signal-dependent Poisson noise if the first quality score and the second quality score are greater than a predetermined threshold score, thereby improving the performance of material discrimination.
Dated October 24, 2024
Arjun Karthik Bala
(IN/PA 1021)
Agent for Applicant