CN117115039A - Poisson diffusion model noise reduction method and system - Google Patents
Poisson diffusion model noise reduction method and system Download PDFInfo
- Publication number
- CN117115039A CN117115039A CN202311164569.7A CN202311164569A CN117115039A CN 117115039 A CN117115039 A CN 117115039A CN 202311164569 A CN202311164569 A CN 202311164569A CN 117115039 A CN117115039 A CN 117115039A
- Authority
- CN
- China
- Prior art keywords
- image
- noise
- gaussian distribution
- reduced
- poisson
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 55
- 230000009467 reduction Effects 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000009466 transformation Effects 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims description 17
- 238000011156 evaluation Methods 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 238000003384 imaging method Methods 0.000 abstract description 10
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000005855 radiation Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000011282 treatment Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 238000002059 diagnostic imaging Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000002603 single-photon emission computed tomography Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention is suitable for the technical field of medical image noise reduction, and provides a poisson diffusion model noise reduction method and system. Compared with the prior art, the invention has the beneficial effects that: the noise reduction diffusion probability model originally applicable to stable variance noise is also applicable to Poisson distribution data of common noise variance along with signal intensity transformation of medical images through interpolation of an image to be reduced after the anscom transformation and the image to be reduced with the lowest quality; the method can adapt to radiation imaging with different noise levels, and noise with different degrees is added in a distributed manner in the forward diffusion process, so that images to be noise-reduced with different noise levels can be recovered through different steps; the method for classifying the noise level of the unknown image to be noise-reduced can acquire the noise level of the image to be noise-reduced, so that the noise level corresponds to proper time sequence parameters, and the image with noise reduction is obtained.
Description
Technical Field
The invention is suitable for the technical field of medical image noise reduction, and particularly relates to a poisson diffusion model noise reduction method and system.
Background
Medical imaging procedures typically have two projection imaging modes: projection imaging and emission imaging. CT is that X-ray passes through the target, and the projection of the imaging object after the X-ray photon absorption is obtained, and SPECT and PET are all obtained by collecting gamma photons emitted by the inside of the target object. Projection imaging, the process of photon distribution from three-dimensional space of an imaged object to detection data, may have some factors between light source emission and CMOS due to the particulation and uncertainty of photons, resulting in that part of photons are not received by CMOS, and this process follows the rule of poisson distribution. Poisson distribution is a distribution with a mean equal to the variance, and although the larger the intensity, the larger the fluctuation of the image, the image signal-to-noise ratio increases with the square root of the number of captured photons. However, the requirements of clinical dosage and acquisition time are limited, the projection image is inevitably accompanied with great noise, and the noise reduction effect of the projection image also becomes an important image factor of the quality of the medical image after final reconstruction.
In recent years, a deep learning diffusion model is a research hot spot, and is mainly applied to the field of image generation at present, and is also gradually applied to various downstream tasks and medical fields. The diffusion model has the advantage of flexibility, and can recover noise with different degrees through the same model. The forward process of the diffusion model gradually loses the recognizable characteristics of the image mainly through small amount of Gaussian noise gradually added, and then gradually reconstructs the image from the noise through the backward diffusion process. However, there has been no noise reduction process applied to medical image imaging for the following reasons: gaussian noise is an additive noise, while poisson noise is a signal dependent noise; the medical image noise reduction should be controlled, and new image features should not be generated, affecting the judgment of the doctor. The current diffusion model based on Gaussian noise is mainly applied in a mode of directly adding Gaussian noise on an original image, but because the difference between the Gaussian noise and the Poisson noise is large, the Gaussian noise still can show larger noise in a region with a lower signal, which is inconsistent with the noise characteristic of a medical image.
Therefore, there is a need for a noise reduction method that satisfies the poisson distribution common to medical images and combines the advantages of diffusion models.
Disclosure of Invention
The invention provides a poisson diffusion model noise reduction method and system, and aims to solve the problem of poisson noise reduction in the medical image imaging process through a diffusion model.
In a first aspect, the present invention provides a poisson diffusion model noise reduction method, the poisson diffusion model noise reduction method including the steps of:
acquiring a plurality of images to be noise reduced;
the noise level of each image to be noise reduced is evaluated according to a preset evaluation method, and corresponding time sequence parameters are obtained;
different processing modes are carried out on the image to be noise reduced according to the size of the time sequence parameter, and a plurality of approximate Gaussian distribution images are obtained;
taking a plurality of the approximate Gaussian distribution images and corresponding time sequence parameters as inputs of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
and carrying out an anscam inverse transformation on the Gaussian distribution image to obtain a poisson distribution image, and outputting the poisson distribution image as a final result.
Preferably, the preset evaluation method includes the following steps:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
Preferably, in the step of performing different processing manners on the image to be noise reduced according to the magnitude of the time sequence parameter to obtain a plurality of approximate gaussian distribution images, the approximate gaussian distribution images include a first approximate gaussian distribution image, a second approximate gaussian distribution image and a third approximate gaussian distribution image, and the processing manners specifically include:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, representing that the noise of the image to be noise-reduced is zero, performing ascombe conversion on the image to be noise-reduced to obtain the first approximate Gaussian distribution image;
when t=t, representing that the noise of the image to be reduced is maximum, generating poisson distribution sampling data corresponding to the average photon count level representing the minimum image quality according to the preset average photon count level, and performing anscam conversion on the poisson distribution sampling data to obtain the second approximate Gaussian distribution image;
when t=1, 2,3, …, T-1, the third approximately gaussian distribution image is obtained by linear weighting of the first and second approximately gaussian distribution images.
Preferably, the calculation formula of the first approximate gaussian distribution image is as follows:
wherein x is 0 Representing the image to be noise reduced without noise,y 0 representing the first approximately gaussian distribution image.
Preferably, the calculation formula of the third approximate gaussian distribution image is as follows:
wherein alpha is t Representing a constant in the (0, 1) interval and increasing with an increase of said timing parameter t, y t Representing the third approximately Gaussian distribution image with the time sequence parameter t, and P-P (x 0 * ) Indicating that p obeys the mean and variance to be x 0 * P represents the image corresponding to the lowest image quality.
In a second aspect, the invention further provides a poisson diffusion model noise reduction system, and an acquisition module is used for acquiring a plurality of images to be noise reduced;
the evaluation module is used for evaluating the noise level of each image to be noise reduced according to a preset evaluation method to obtain corresponding time sequence parameters;
the processing module is used for carrying out different processing modes on the image to be noise reduced according to the time sequence parameters to obtain a plurality of approximate Gaussian distribution images;
the input module is used for taking a plurality of the approximate Gaussian distribution images and corresponding time sequence parameters as the input of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
and the inverse transformation module is used for carrying out an anscam inverse transformation on the Gaussian distribution image to obtain a poisson distribution image, and outputting the poisson distribution image as a final result.
Preferably, the preset evaluation method includes the following steps:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
Preferably, in the step of performing different processing manners on the image to be noise reduced according to the magnitude of the time sequence parameter to obtain a plurality of approximate gaussian distribution images, the approximate gaussian distribution images include a first approximate gaussian distribution image, a second approximate gaussian distribution image and a third approximate gaussian distribution image, and the processing manners specifically include:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, representing that the noise of the image to be noise-reduced is zero, performing ascombe conversion on the image to be noise-reduced to obtain the first approximate Gaussian distribution image;
when t=t, representing that the noise of the image to be reduced is maximum, generating poisson distribution sampling data corresponding to the average photon count level representing the minimum image quality according to the preset average photon count level, and performing anscam conversion on the poisson distribution sampling data to obtain the second approximate Gaussian distribution image;
when t=1, 2,3, …, T-1, the third approximately gaussian distribution image is obtained by linear weighting of the first and second approximately gaussian distribution images.
Compared with the prior art, the invention has the beneficial effects that: the noise reduction diffusion probability model originally applicable to stable variance noise is also applicable to Poisson distribution data of common noise variance along with signal intensity transformation of medical images through interpolation of an image to be reduced after the anscom transformation and the image to be reduced with the lowest quality; the method can adapt to radiation imaging with different noise levels, and noise with different degrees is added in a distributed manner in the forward diffusion process, so that images to be noise-reduced with different noise levels can be recovered through different steps; the method for classifying the noise level of the unknown image to be noise-reduced can acquire the noise level of the image to be noise-reduced, so that the noise level corresponds to proper time sequence parameters, and the image with noise reduction is obtained.
Drawings
The present invention will be described in detail with reference to the accompanying drawings. The foregoing and other aspects of the invention will become more apparent and more readily appreciated from the following detailed description taken in conjunction with the accompanying drawings. In the accompanying drawings:
FIG. 1 is a block flow diagram of a poisson diffusion model denoising method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of functional partitioning of a Poisson diffusion model noise reduction method according to an embodiment of the present invention;
FIG. 3 is an Anscombe transform schematic diagram of a Poisson diffusion model denoising method according to an embodiment of the present invention;
FIG. 4 is a schematic view of medical images after different treatments of the Poisson diffusion model denoising method according to an embodiment of the present invention;
fig. 5 is a flowchart of a poisson diffusion model noise reduction system according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Example one
Referring to fig. 1-4, the present invention provides a poisson diffusion model noise reduction method, which includes the following steps:
s101, acquiring a plurality of images to be noise reduced;
in the embodiment of the invention, the image to be noise-reduced refers to an image needing noise reduction in a real medical scene.
S102, evaluating the noise level of each image to be noise reduced according to a preset evaluation method to obtain corresponding time sequence parameters;
in an embodiment of the present invention, the preset evaluation method includes the following steps:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
S103, carrying out different processing modes on the image to be noise reduced according to the size of the time sequence parameter to obtain a plurality of approximate Gaussian distribution images;
in the embodiment of the present invention, referring to fig. 3, the approximately gaussian distribution image includes a first approximately gaussian distribution image, a second approximately gaussian distribution image, and a third approximately gaussian distribution image, and the processing method specifically includes:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, representing that the noise of the image to be denoised is zero, performing an ascombe conversion on the image to be denoised to obtain the first approximate gaussian distribution image y 0 The method comprises the steps of carrying out a first treatment on the surface of the The first approximately Gaussian distribution image y 0 The calculation formula of (2) is as follows:
wherein x is 0 Representing the noise-free image to be noise-reduced, y 0 Representing the first approximate GaussianThe image is distributed.
When t=t, representing that the noise of the image to be reduced is maximum, generating poisson distribution sampling data corresponding to the average photon count level m representing the minimum image quality according to the preset average photon count level m, and performing anscombe conversion on the poisson distribution sampling data to obtain the second approximate gaussian distribution image y T The method comprises the steps of carrying out a first treatment on the surface of the Photon counting refers to the number of photons of X-rays or the number of gamma photons in obtaining medical imaging. In addition, the probability density function of the poisson distribution represents the probability distribution of the occurrence times of photon events in unit time, and the expression is as follows:
where k represents the occurrence of photon time per unit time, λ represents the average occurrence of photon events per unit time, and poisson distribution is a distribution with equal mean and variance. X is x T ～P( 0 * ) Represents x T Obeying the mean and variance to be x 0 * Poisson distribution of (a).
When t=1, 2,3, …, T-1, the first approximately gaussian distribution image y is passed through 0 And the second approximately Gaussian distribution image y T Is then applied to the image of the third approximate gaussian distribution.
The calculation formula of the third approximate Gaussian distribution image is as follows:
wherein alpha is t Representing a constant in the (0, 1) interval and increasing with an increase of said timing parameter t, y t Representing the third approximately Gaussian distribution image with the time sequence parameter t, and P-P (x 0 * ) Indicating that p obeys the mean and variance to be x 0 * P represents the image corresponding to the lowest image quality.
Specifically, α t Is set of (1)The sum may be a linear uniform division of 0-1, or a division subject to a sigmoid function or a Cosine function distribution. Alpha t The change in the magnitude of the parameter t over time is shown in fig. 2, where Linear represents the Linear division of equally divided.
S104, taking a plurality of approximate Gaussian distribution images and corresponding time sequence parameters as inputs of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
in the embodiment of the invention, the trained noise reduction and diffusion probability model is a U-Net network with an attention mechanism and time position coding based on a network frame, and model parameters are optimized by adopting an L1 loss function. The noise reduction diffusion probability model gradually adds Gaussian-distributed noise to the approximately Gaussian-distributed image.
S105, carrying out an anscam inverse transformation on the Gaussian distribution image to obtain a poisson distribution image, and outputting the poisson distribution image as a final result.
In the embodiment of the present invention, referring to fig. 4, in fig. 4, the first line of pictures represents the gaussian distribution image generated in the above step, the second line of pictures represents the poisson distribution image obtained after the inverse transformation, and the third line of pictures is poisson distribution adoption data conforming to the noise level of the real data.
Compared with the prior art, the invention has the beneficial effects that: the noise reduction diffusion probability model originally applicable to stable variance noise is also applicable to Poisson distribution data of common noise variance along with signal intensity transformation of medical images through interpolation of an image to be reduced after the anscom transformation and the image to be reduced with the lowest quality; the method can adapt to radiation imaging with different noise levels, and noise with different degrees is added in a distributed manner in the forward diffusion process, so that images to be noise-reduced with different noise levels can be recovered through different steps; the method for classifying the noise level of the unknown image to be noise-reduced can acquire the noise level of the image to be noise-reduced, so that the noise level corresponds to proper time sequence parameters, and the image with noise reduction is obtained.
Example two
Referring to fig. 5, the present invention further provides a poisson diffusion model noise reduction system 200, which includes:
s201, an acquisition module is used for acquiring a plurality of images to be noise reduced;
in the embodiment of the invention, the image to be noise-reduced refers to an image needing noise reduction in a real medical scene.
S202, an evaluation module is used for evaluating the noise level of each image to be noise reduced according to a preset evaluation method to obtain corresponding time sequence parameters;
in an embodiment of the present invention, the preset evaluation method includes the following steps:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
S203, a processing module is used for carrying out different processing modes on the image to be noise reduced according to the size of the time sequence parameter to obtain a plurality of approximate Gaussian distribution images;
in the embodiment of the present invention, referring to fig. 3, the approximately gaussian distribution image includes a first approximately gaussian distribution image, a second approximately gaussian distribution image, and a third approximately gaussian distribution image, and the processing method specifically includes:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, the noise of the image to be noise reduced is representedThe sound is zero, and the image to be noise reduced is subjected to anscam conversion to obtain the first approximate Gaussian distribution image y 0 The method comprises the steps of carrying out a first treatment on the surface of the The first approximately Gaussian distribution image y 0 The calculation formula of (2) is as follows:
wherein x is 0 Representing the noise-free image to be noise-reduced, y 0 Representing the first approximately gaussian distribution image.
When t=t, the noise representing the image to be noise reduced is the maximum value, and poisson distribution sampling data x corresponding to the average photon count level m representing the minimum image quality is generated according to the preset average photon count level m T Sampling data x for the poisson distribution T Performing an anscom conversion to obtain the second approximate Gaussian distribution image y T The method comprises the steps of carrying out a first treatment on the surface of the Photon counting refers to the number of photons of X-rays or the number of gamma photons in obtaining medical imaging. In addition, the probability density function of the poisson distribution represents the probability distribution of the occurrence times of photon events in unit time, and the expression is as follows:
where k represents the occurrence of photon time per unit time, λ represents the average occurrence of photon events per unit time, and poisson distribution is a distribution with equal mean and variance. X is x T ～P( 0 * ) Represents x T Obeying the mean and variance to be x 0 * Poisson distribution of (a).
When t=1, 2,3, …, T-1, the first approximately gaussian distribution image y is passed through 0 And the second approximately Gaussian distribution image y T Is then applied to the image of the third approximate gaussian distribution.
The calculation formula of the third approximate Gaussian distribution image is as follows:
wherein alpha is t Representing a constant in the (0, 1) interval and increasing with an increase of said timing parameter t, y t Representing the third approximately Gaussian distribution image with the time sequence parameter t, and P-P (x 0 * ) Indicating that p obeys the mean and variance to be x 0 * P represents the image corresponding to the lowest image quality.
Specifically, α t The set of (a) may be a linear uniform division of 0-1 or a division subject to a sigmoid function or a Cosine function distribution. Alpha t The change in the magnitude of the parameter t over time is shown in fig. 2, where Linear represents the Linear division of equally divided.
S204, an input module takes a plurality of approximate Gaussian distribution images and corresponding time sequence parameters as inputs of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
in the embodiment of the invention, the trained noise reduction and diffusion probability model is a U-Net network with an attention mechanism and time position coding based on a network frame, and model parameters are optimized by adopting an L1 loss function. The noise reduction diffusion probability model gradually adds Gaussian-distributed noise to the approximately Gaussian-distributed image.
S205, performing an anscam inverse transformation on the Gaussian distribution image to obtain a Poisson distribution image, and outputting the Poisson distribution image as a final result.
In the embodiment of the present invention, referring to fig. 4, in fig. 4, the first line of pictures represents the gaussian distribution image generated in the above step, the second line of pictures represents the poisson distribution image obtained after the inverse transformation, and the third line of pictures is poisson distribution adoption data conforming to the noise level of the real data.
The poisson diffusion model noise reduction system 200 can implement the steps in the poisson diffusion model noise reduction method in the above embodiment, and can achieve the same technical effects, and is not described herein again with reference to the description in the above embodiment.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
While the embodiments of the present invention have been illustrated and described in connection with the drawings, what is presently considered to be the most practical and preferred embodiments of the invention, it is to be understood that the invention is not limited to the disclosed embodiments, but on the contrary, is intended to cover various equivalent modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims (8)
1. The poisson diffusion model noise reduction method is characterized by comprising the following steps of:
acquiring a plurality of images to be noise reduced;
the noise level of each image to be noise reduced is evaluated according to a preset evaluation method, and corresponding time sequence parameters are obtained;
different processing modes are carried out on the image to be noise reduced according to the size of the time sequence parameter, and a plurality of approximate Gaussian distribution images are obtained;
taking a plurality of the approximate Gaussian distribution images and corresponding time sequence parameters as inputs of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
and carrying out an anscam inverse transformation on the Gaussian distribution image to obtain a poisson distribution image, and outputting the poisson distribution image as a final result.
2. The poisson diffusion model noise reduction method according to claim 1, wherein the preset evaluation method comprises the steps of:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
3. The poisson diffusion model denoising method according to claim 1, wherein in the step of obtaining a plurality of approximately gaussian distribution images by performing different processing modes on the image to be denoised according to the magnitude of the time sequence parameter, the approximately gaussian distribution images include a first approximately gaussian distribution image, a second approximately gaussian distribution image and a third approximately gaussian distribution image, and the processing modes are specifically as follows:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, representing that the noise of the image to be noise-reduced is zero, performing ascombe conversion on the image to be noise-reduced to obtain the first approximate Gaussian distribution image;
when t=t, representing that the noise of the image to be reduced is maximum, generating poisson distribution sampling data corresponding to the average photon count level representing the minimum image quality according to the preset average photon count level, and performing anscam conversion on the poisson distribution sampling data to obtain the second approximate Gaussian distribution image;
when t=1, 2,3, …, T-1, the third approximately gaussian distribution image is obtained by linear weighting of the first and second approximately gaussian distribution images.
4. The poisson diffusion model noise reduction method according to claim 3, wherein the first approximately gaussian distribution image is calculated as follows:
wherein x is 0 Representing the noise-free image to be noise-reduced, y 0 Representing the first approximately gaussian distribution image.
5. The poisson diffusion model noise reduction method according to claim 4, wherein the third approximate gaussian distribution image is calculated as follows:
wherein alpha is t Representing a constant in the (0, 1) interval and increasing with an increase of said timing parameter t, y t Representing the third approximately Gaussian distribution image with the time sequence parameter t, and P-P (x 0 * η) represents that p obeys the mean and variance to be x 0 * And the Poisson distribution of eta, and p represents the image corresponding to the lowest image quality.
6. A poisson diffusion model noise reduction system, the poisson diffusion model noise reduction system comprising:
the acquisition module is used for acquiring a plurality of images to be noise reduced;
the evaluation module is used for evaluating the noise level of each image to be noise reduced according to a preset evaluation method to obtain corresponding time sequence parameters;
the processing module is used for carrying out different processing modes on the image to be noise reduced according to the time sequence parameters to obtain a plurality of approximate Gaussian distribution images;
the input module is used for taking a plurality of the approximate Gaussian distribution images and corresponding time sequence parameters as the input of a trained noise reduction diffusion probability model to obtain Gaussian distribution images; the noise reduction and diffusion probability model is provided with T time steps, and each time step is provided with the corresponding time sequence parameter;
and the inverse transformation module is used for carrying out an anscam inverse transformation on the Gaussian distribution image to obtain a poisson distribution image, and outputting the poisson distribution image as a final result.
7. The poisson diffusion model noise reduction system according to claim 6, wherein the preset evaluation method comprises the steps of:
dividing the image to be noise reduced into N rectangular areas according to the width and the height of the image to be noise reduced;
calculating the variances of photon counts of N rectangular areas respectively, wherein the position of the photon count of 0 is not counted into calculation;
counting the maximum value and the minimum value of the variance in the N rectangular areas;
estimating the noise level of the current image to be noise reduced according to the maximum value of the variance divided by the minimum value of the variance;
and obtaining the corresponding time sequence parameter according to the corresponding relation between the current noise level of the image to be reduced and the time sequence parameter through a preset noise level numerical range.
8. The poisson diffusion model noise reduction system according to claim 7, wherein in the step of obtaining a plurality of approximately gaussian distribution images by performing different processing manners on the image to be noise reduced according to the magnitude of the time sequence parameter, the approximately gaussian distribution images include a first approximately gaussian distribution image, a second approximately gaussian distribution image, and a third approximately gaussian distribution image, and the processing manners are specifically:
defining the time sequence parameter as T, wherein T is more than or equal to 0 and less than or equal to T;
when t=0, representing that the noise of the image to be noise-reduced is zero, performing ascombe conversion on the image to be noise-reduced to obtain the first approximate Gaussian distribution image;
when t=t, representing that the noise of the image to be reduced is maximum, generating poisson distribution sampling data corresponding to the average photon count level representing the minimum image quality according to the preset average photon count level, and performing anscam conversion on the poisson distribution sampling data to obtain the second approximate Gaussian distribution image;
when t=1, 2,3, …, T-1, the third approximately gaussian distribution image is obtained by linear weighting of the first and second approximately gaussian distribution images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311164569.7A CN117115039A (en) | 2023-09-08 | 2023-09-08 | Poisson diffusion model noise reduction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311164569.7A CN117115039A (en) | 2023-09-08 | 2023-09-08 | Poisson diffusion model noise reduction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117115039A true CN117115039A (en) | 2023-11-24 |
Family
ID=88810890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311164569.7A Pending CN117115039A (en) | 2023-09-08 | 2023-09-08 | Poisson diffusion model noise reduction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117115039A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117614467A (en) * | 2024-01-17 | 2024-02-27 | 青岛科技大学 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
-
2023
- 2023-09-08 CN CN202311164569.7A patent/CN117115039A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117614467A (en) * | 2024-01-17 | 2024-02-27 | 青岛科技大学 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
| CN117614467B (en) * | 2024-01-17 | 2024-05-07 | 青岛科技大学 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097611B (en) | Image reconstruction method, device, equipment and storage medium | |
| Chen et al. | Nonlocal prior Bayesian tomographic reconstruction | |
| US8058601B2 (en) | Determining a multimodal pixon map for tomographic-image reconstruction | |
| CN112348936B (en) | Low-dose cone-beam CT image reconstruction method based on deep learning | |
| KR20130038794A (en) | Method of noise reduction in digital x-ray frames series | |
| US10832381B2 (en) | Processing at least one X-ray image | |
| Hasan et al. | Hybrid-collaborative Noise2Noise denoiser for low-dose CT images | |
| Fang et al. | Removing ring artefacts for photon-counting detectors using neural networks in different domains | |
| CN117115039A (en) | Poisson diffusion model noise reduction method and system | |
| CN111353958A (en) | Image processing method, device and system | |
| JP2012150802A (en) | Noise assessment method for digital x-ray films | |
| CN114387359A (en) | Three-dimensional X-ray low-dose imaging method and device | |
| Yu et al. | Noise type evaluation in positron emission tomography images | |
| CN111260583A (en) | Multi-discriminant-based multi-analysis network missing CT projection data estimation method | |
| CN111968192A (en) | Construction method of CT image, CT device and storage medium | |
| Chang et al. | Statistical x-ray computed tomography imaging from photon-starved measurements | |
| CN116342414A (en) | CT image noise reduction method and system based on similar block learning | |
| Carasso | Error bounds in nonsmooth image deblurring | |
| Lu et al. | Adaptive Image De-noising Method Based on Spatial Autocorrelation | |
| JP6974159B2 (en) | Image processing device and image processing method | |
| Mahboob et al. | A Novel Medical Image De-noising Algorithm for Efficient Diagnosis in Smart Health Environment | |
| Choi et al. | Statistical image restoration for low-dose CT using convolutional neural networks | |
| Zubair et al. | Enhancing Low-Dose CT Image Quality Through Deep Learning: A DoG-Sharpened U-Net Approach With Attention Mechanism | |
| Liang et al. | A model-based deep learning reconstruction for X-Ray CT | |
| JP6986961B2 (en) | Image processing device and image processing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |