Full-dynamic illumination real-time medical image three-dimensional body rendering method and computer program product

Abstract

The invention discloses a full-dynamic illumination real-time medical image three-dimensional body rendering method and a computer program product, and relates to the technical field of computer vision and computer imageology. The full-dynamic illumination real-time medical image three-dimensional volume rendering method comprises the following steps: acquiring volume data, wherein the volume data is three-dimensional information corresponding to a three-dimensional object; performing coloring estimation on the volume data based on a Monte Carlo algorithm, and generating a radiation sample; feature decoupling is conducted on the radiation sample, so that the radiation sample is decomposed into a low-detail high-noise assembly and a high-detail low-noise assembly, the low-detail high-noise assembly comprises light source radiation and light transmissivity related information, and the high-detail low-noise assembly comprises scattering attributes related to surface materials; denoising the low-detail high-noise component and the high-detail low-noise component respectively; and combining the de-noised low-detail high-noise component and the de-noised high-detail low-noise component, and generating a de-noised image. By adopting the technology provided by the invention, the image rendering quality can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and computer graphics, and in particular to a real-time medical image three-dimensional volume rendering method with full dynamic illumination and a computer program product. Background Art

[0002] Medical imaging 3D volume rendering technology generates 3D display effects through visualization technology from data scanned by medical imaging devices such as CT and MRI (magnetic resonance imaging), achieving detailed visualization of human organs, bones and other structures, allowing doctors to observe and analyze the internal structure of the human body from various angles, greatly assisting disease diagnosis and treatment planning.

[0003] At present, in the field of medical imaging, realistic rendering technology is mainly used to improve the realism of images to enhance doctors' ability to interpret image data. By reproducing complex light and shadow effects and real material properties, doctors can observe more subtle structural differences from images, such as the density and texture of different tissues. This advanced visual presentation can help doctors more accurately identify lesion areas, deeply understand the spatial distribution and impact of the disease, and make more accurate judgments in treatment planning.

[0004] However, although photorealistic rendering can provide remarkable visual effects, it still faces many challenges in real-time 3D volume rendering applications. In particular, high noise levels are a common problem in real-time rendering, which may mask important details and reduce the diagnostic quality of images. In 3D volume rendering of medical images, high fidelity and rich details are crucial, and conventional denoising techniques often cannot effectively reduce noise without losing these details.

[0005] Based on this, the Chinese invention patent document (CN119205549A) discloses a medical image denoising model training method and device. By combining the deep Hessian attention feature, the U-Net network's attention to structural details is enhanced. When removing the inherent speckle of medical images and reconstructing the original medical tissue structure, more attention is paid to the tissue boundary information and texture details in the medical image, which helps tissue stratification, thereby improving the image quality of medical images after denoising. However, it still needs further improvement in terms of real-time interactivity and user perception. Summary of the invention

[0006] The present invention provides a real-time medical image three-dimensional volume rendering method with full dynamic illumination, so as to solve the problem of low real-time interactivity of the three-dimensional volume rendering technology in the prior art.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is to provide a real-time medical image 3D volume rendering method with full dynamic lighting, and the real-time medical image 3D volume rendering method with full dynamic lighting comprises: Acquire volume data, which is three-dimensional information corresponding to a three-dimensional object; perform coloring estimation on the volume data based on a Monte Carlo algorithm and generate radiation samples; perform feature decoupling on the radiation samples to decompose them into a low-detail high-noise component and a high-detail low-noise component, wherein the low-detail high-noise component contains information related to light source radiation and light transmittance, and the high-detail low-noise component contains scattering properties related to surface material; denoise the low-detail high-noise component and the high-detail low-noise component respectively; combine the denoised low-detail high-noise component and the high-detail low-noise component, and generate a denoised image.

[0008] In some embodiments, performing coloring estimation on the volume data based on a Monte Carlo algorithm and generating radiation samples includes: A surface medium coexistence shading model is pre-constructed, wherein the surface medium coexistence shading model defines a composite particle having both body scattering characteristics and surface scattering characteristics; the scattering distribution function of the composite particle is defined as:

[0009] in, represents the BRDF value, represents the volume scattering probability at the sampling point x, represents the surface scattering probability at the sampling point x, represents the phase function.

[0010] In some embodiments, the method of performing shading estimation on the volume data based on the Monte Carlo algorithm and generating radiation samples further includes: performing direct illumination estimation on the volume data based on the surface medium coexistence shading model, which is expressed as:

[0011] in, ) represents the sampling point x to the sampling point The light transmittance between Represents the sampling point x The light intensity in a direction, Indicates that the light source is at the sampling point The sampling density of .

[0012] In some embodiments, the radiation sample is feature decoupled to decompose it into a low-detail high-noise component and a high-detail low-noise component, including: decomposing the radiation sample into the low-detail high-noise component and the high detail low noise components , recorded as:

[0013]

[0014] in, represents the radiation of the light source, represents the refractive index of light, Indicates that the light source is at the sampling point The sampling density of Represents the lighting estimation result.

[0015] In some embodiments, the denoising the low detail high noise component and the high detail low noise component separately further comprises: Determine the first real scattering event, when the particle density of the sampling point of the real scattering event exceeds a preset threshold, record the depth value of the corresponding sampling point; select the minimum depth value among the recorded depth values ​​of the corresponding sampling points, and use the minimum depth value for spatial denoising of the high-detail low-noise component and / or spatial denoising of the low-detail high-noise component; when the particle density of all sampling points is lower than the preset threshold, select the middle value of the depth values ​​of all sampling points, and use the middle depth value for spatial denoising of the high-detail low-noise component and / or spatial denoising of the low-detail high-noise component.

[0016] In some embodiments, the denoising the low detail high noise component and the high detail low noise component respectively comprises: High detail low noise components Perform spatial denoising; for the high detail low noise components Performing time series denoising, wherein the time series denoising includes: calculating the distance of the geometric auxiliary features of the relevant pixels ; The distance of the geometric auxiliary feature of the relevant pixel in exponential form As the historical frame weight; the current frame weight is recorded as: ; Blend the rendering result of the previous frame into the current frame.

[0017] In some embodiments, the high detail low noise component Denoising also includes performing spatiotemporal denoising on the opacity of the pixels in the current frame.

[0018] In some embodiments, the denoising of the low detail high noise component and the high detail low noise component respectively includes: using the denoised high detail low noise component as an auxiliary feature to guide the spatial denoising of the low detail high noise component.

[0019] In some embodiments, the method of performing coloring estimation on the volume data based on a Monte Carlo algorithm and generating radiation samples further includes: applying a hybrid acceleration structure to sample the volume data, specifically: The three-dimensional space of the volume data is divided into a plurality of macro units, non-empty macro units are determined, and an octree is constructed to recursively divide the macro units into a plurality of octree nodes; the octree nodes are linearly stored in spatial order, wherein each node contains an index of a parent node, indexes of all child nodes, and a range of values ​​of voxels in a spatial region represented by most octree nodes.

[0020] The technical solution provided by the present invention has the following beneficial effects compared with the prior art: The hybrid acceleration structure using macro units and octrees can optimize sampling efficiency. Combined with the surface medium coexistence shading model to optimize shading estimation, low-noise rendering results can be quickly obtained, thereby improving real-time interactivity and providing medical staff with more accurate and realistic images, enhancing their visual experience so that they can make more accurate diagnoses.

[0021] Among them, the radiation samples after shading estimation are feature decoupled to decompose them into low-detail high-noise components and high-detail low-noise components, so as to ensure the details of the rendering results while removing noise, thereby effectively improving the quality of image rendering of three-dimensional objects, ensuring real-time denoising effect while retaining image details, and finally, by combining the denoised low-detail high-noise components with the high-detail low-noise components, the blurring effect can be reduced, making the rendering results more realistic and expressive.

[0022] In some embodiments, the present application also provides a computer program product, which includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the above-mentioned real-time medical image three-dimensional volume rendering method with full dynamic lighting is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 The present invention provides a method for real-time 3D medical image rendering with full dynamic illumination. Figure 1 ; Figure 2 It is an algorithm block diagram of an embodiment of a method for real-time medical image three-dimensional volume rendering with full dynamic illumination provided by the present invention; Figure 3 This is a comparison diagram of the coloring effects of an embodiment of a real-time medical image three-dimensional volume rendering method with full dynamic illumination provided by the present invention; Figure 4 The present invention provides a method for real-time 3D medical image rendering with full dynamic illumination. Figure 2 ; Figure 5 It is a space partition diagram of a macrogrid and octree hybrid acceleration structure of a real-time medical image three-dimensional volume rendering method with full dynamic illumination provided by the present invention. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 to Figure 2 As shown, Figure 1 The present invention provides a method for real-time 3D medical image rendering with full dynamic illumination. Figure 1 ; Figure 2 An algorithm block diagram of an embodiment of a real-time medical image 3D volume rendering method with full dynamic illumination provided by the present application is shown.

[0026] In some embodiments, a method for real-time medical image 3D volume rendering with full dynamic lighting includes: Step S100: acquiring volume data, where the volume data is three-dimensional information corresponding to a three-dimensional object.

[0027] Volume data can be obtained through medical scanning (such as CT or MRI), 3D scanning or other 3D data acquisition technology. For example, volume data may include the geometry of the 3D object, attribute information, such as Figure 2 Density, color, and material properties of the skull areas shown.

[0028] Step S200 , performing coloring estimation on volume data based on a Monte Carlo algorithm and generating radiation samples.

[0029] In order to render the three-dimensional volume of medical images more realistically and simulate the interaction between light and matter, so as to provide more accurate and realistic images visually, in the realistic volume rendering of medical images, it is usually necessary to reveal the implicit surface structure in the volume data to enhance the authenticity and expressiveness of the image. Among them, the Monte Carlo algorithm is a numerical calculation method based on random sampling. The volume path tracing process of applying it to realistic rendering of images can be summarized as the following steps: (1) Sampling free path length: Determine the path length of light propagating in the scene through random sampling to simulate the propagation of light in space.

[0030] (2) Calculate direct illumination at real scattering locations: Calculate the interaction between light and three-dimensional objects, including reflection, refraction, and scattering of light.

[0031] (3) Sampling new scattering directions and updating path transmittance: The new scattering directions are sampled to determine the next propagation direction of the light, and the path transmittance is updated to simulate the attenuation effect of light when passing through different media.

[0032] Step S300, feature decoupling is performed on the radiation sample to decompose it into a low-detail high-noise component and a high-detail low-noise component, wherein the low-detail high-noise component contains information related to the radiation of the light source and the transmittance of the light, and the high-detail low-noise component contains scattering properties related to the surface material.

[0033] In surface model rendering, the radiation estimated using the Monte Carlo method can be described as the product of reflectivity and illumination. Since the intersection of the sampling ray and the surface of the three-dimensional object is certain (the calculation of reflectivity is based on accurate geometric features rather than random sampling), the reflectivity at the surface intersection can be considered to be noise-free, so only the illumination component needs to be denoised.

[0034] However, in the prior art, when using the Monte Carlo algorithm to estimate radiation, overall denoising is usually used, and the denoising is guided by auxiliary features with noise, thereby introducing a blurring effect (causing the image details in the rendering result to become blurred and lose clarity and sharpness). Through step S300, decoupling denoising can be used to separate the features of the volume rendered radiation samples (shading estimation results), thereby achieving noise removal while ensuring the detailed features of the rendered structure.

[0035] Step S400, denoising the low-detail high-noise component and the high-detail low-noise component respectively.

[0036] By decomposing the radiation samples into low detail high noise (LDHN) components and high detail low noise (HDLN) components, LDHN is related to illumination estimation, such as light source radiation and light transmittance; HDLN is related to surface material, such as surface scattering property information of three-dimensional objects. By decomposing the radiation samples for independent denoising, it is possible to more effectively preserve image details while reducing noise. Exemplarily, HDLN is used to preserve image details while also providing auxiliary guidance to LDHN.

[0037] Step S500, combining the denoised low-detail high-noise component and the high-detail low-noise component to generate a denoised image.

[0038] Exemplarily, the independently denoised low detail high noise component and the high detail low noise component are multiplied to merge the information of the two components, wherein the low detail high noise component provides the lighting and color information of the image, and the high detail low noise component provides the detail and texture information of the three-dimensional object.

[0039] In the embodiment of the present application, the above-mentioned realistic rendering method can more realistically simulate the interaction between three-dimensional objects in the light domain, thereby providing medical staff with more accurate and realistic images visually. Among them, by performing feature decoupling on the radiation samples after shading estimation, it is decomposed into low-detail high-noise components and high-detail low-noise components, so as to ensure the details of the rendering results while removing noise, thereby effectively improving the quality of image rendering of three-dimensional objects.

[0040] In some embodiments, step S200, performing shading estimation on volume data based on a Monte Carlo algorithm and generating radiation samples, includes: pre-constructing a surface-medium coexistence shading model, wherein the surface-medium coexistence shading model defines a composite particle having both volume scattering characteristics and surface scattering characteristics.

[0041] The scattering distribution function of the composite particle is defined as:

[0042] in, represents the BRDF value, represents the volume scattering probability at the sampling point x, represents the surface scattering probability at the sampling point x, represents the phase function.

[0043] In an embodiment of the present application, when rendering a three-dimensional volume with multiple materials, two main types of particles can be defined in the volume space (the space for storing and processing three-dimensional data). One type is reflection particles used to describe surface reflection characteristics (simulating the light reflection effect from the implicit surface in the volume data), and the other type is volume scattering particles used to describe volume scattering characteristics (simulating the light scattering effect on the surface of the sampling point). The two types of particles are used to simulate the optical behavior of complex materials, thereby achieving high-quality rendering effects.

[0044] For example, the current technique for estimating direct illumination by importance sampling can be expressed as:

[0045] in, Represents the sampling point x to the sampling point The light transmittance between Represents the sampling point x The light intensity in a direction, Indicates that the light source is at the sampling point The sampling density of . represents the volume scattering probability at the sampling point x, represents the surface scattering probability at the sampling point x, represents the phase function. Represents the current general lighting estimation result.

[0046] When a scattering event occurs at the sampling point x, it is necessary to determine the type of particle that the light collides with. Specifically, (1) a uniform random number between 0 and 1 is selected. , (2) If the random number Less than , then the collision is considered to occur on the surface scattering particles, recorded as the event in the above formula , if the random number Greater than , then the collision is considered to occur in the volume scattering particle, recorded as the event in the above formula .

[0047] However, the above method needs to distinguish the type of each particle, which leads to low rendering efficiency and low real-time interactivity. Therefore, in step S200, a composite particle with both solid scattering characteristics and surface scattering characteristics is defined by pre-constructing a surface medium coexistence shading model; a single particle can contain two scattering models, so that when a real scattering event occurs, there is no need to randomly select the type of colliding particles, so when a real scattering event occurs, the Monte Carlo algorithm does not need to randomly select the type of colliding particles, and can perform direct illumination estimation on the volume data based on the surface medium coexistence shading model, which is recorded as:

[0048] in, Represents the sampling point x to the sampling point The light transmittance between Represents the sampling point x The light intensity in a direction, Indicates that the light source is at the sampling point The sampling density of .

[0049] For example, in combination Figure 3 As shown, Figure 3 The figure shows the comparison of shading effects. The first column shows the shading effect of defining volume scattering particles and reflective particles, and the second column shows the shading effect based on the surface medium coexistence shading model. The first column and the second column use the same number of samples (1024 samples and 3 samples, respectively). The peak signal-to-noise ratio (PSNR) of the first column is 7.31 and 6.87, and the peak signal-to-noise ratio (PSNR) of the second column is 10.12 and 9.82. Since the PSNR value can be used to measure image quality, the higher the value, the better the image quality. Figure 3 As shown, it can be seen that when the number of sampling samples is larger, the image quality is better, and the shading rendering quality of the image is better when the number of sampling samples is the same using the above-mentioned surface medium coexistence shading model.

[0050] Compared with the combination of bidirectional reflectance distribution function (BRDF) and phase function, the above-mentioned surface medium coexistence model can show smaller estimation variance in the shading estimation process, that is, it is superior in the stability of shading results, and the shading results in different application scenarios are more consistent and less volatile.

[0051] Moreover, the surface medium coexistence model is unbiased to ensure the accuracy and stability of the final rendering results. It can enhance the authenticity and expressiveness of the image by revealing the implicit surface structure in the volume data, and assist medical staff in making more accurate judgments on the current status of three-dimensional objects.

[0052] Combination Figure 4 As shown, Figure 4 The present invention provides a method for real-time 3D medical image rendering with full dynamic illumination. Figure 2 .

[0053] In some embodiments, step S300, feature decoupling of the radiation sample to decompose it into a low-detail high-noise component and a high-detail low-noise component, includes: Step S310, decomposing the radiation sample into low detail high noise components and high detail low noise components , recorded as:

[0054]

[0055] in, represents the radiation of the light source, represents the refractive index of light, Indicates that the light source is at the sampling point The sampling density of Represents the lighting estimation result.

[0056] In order to further reduce noise and enhance the temporal stability of volume rendering results of 3D objects, the radiation samples are decoupled into two components based on feature decoupling, and denoised using corresponding spatiotemporal denoising. The low-detail high-noise component contains the noise of the image (the noise may come from insufficient number of samples or sampling randomness), and the high-detail low-noise component contains the main detail information in the image.

[0057] For example, although reprojection is not the most accurate method to obtain time-correlated pixels, its computational speed enables it to ensure real-time interactivity in real-time rendering. Since volume path tracing sampling is limited by insufficient sampling density and noise introduced by random sampling, it may be impossible to determine the unique intersection in the volume space. Currently, the distance from the camera origin to the position of the first scattering event is used as the depth value for reprojection. However, when there is a large area of ​​low particle density in the volume space, the reprojection strategy using this depth value will be less accurate.

[0058] Therefore, in some embodiments, step S300, denoising the low detail high noise component and the high detail low noise component respectively, further includes: Determine the first real scattering event. When the particle density of the sampling point of the real scattering event exceeds the preset threshold, record the depth value of the corresponding sampling point. When multiple samples are sampled, select the minimum depth value among the recorded depth values ​​of the corresponding sampling points for the reprojection strategy. That is, use the minimum depth value for high-detail low-noise components. Spatial denoising and low detail high noise components Spatial denoising.

[0059] When the particle density of all sampling points is lower than the preset threshold, the middle value of the depth values ​​of all sampling points is selected and the middle depth value is used for high-detail low-noise components Spatial denoising and low detail high noise components The spatial denoising is performed. As a result, the shading results at locations with high particle density contribute more to the image and are more representative of the pixel value of the pixel. When the particle density is low, there are also locations that can be used for reprojection. Using the middle value of the depth value (the median of the depth values ​​of multiple sampling points) instead of the average value can avoid the influence of outliers on reprojection.

[0060] For example, the preset threshold may be one quarter of the maximum particle density in the volume space, and the reprojection formula is recorded as:

[0061] Among them, the pixel j of the current frame is reprojected to the pixel i of the previous frame.

[0062] Step S320, for high detail low noise components Perform spatial denoising, specifically: use depth values ​​and reflectivity at scattering points as auxiliary features to denoise high-detail low-noise components Perform joint bilateral filtering. Joint bilateral filtering is a nonlinear filtering technique that reduces noise while maintaining edges. Since the noise level of high-detail low-noise components is low, joint bilateral filtering can be used as an auxiliary feature in spatial denoising (for example, one joint bilateral filtering).

[0063] In the filtering process, the depth value and the reflectivity at the scattering point (i.e., the reflectivity of light at that point) are used as auxiliary features. For example, first, the temporally related pixels are found through temporal reprojection, and then the temporal weight is calculated based on the similarity of the geometric auxiliary features of the current frame result and the previous frame result, including calculating the distance of the geometric auxiliary features of the related pixels. , and then use the exponential form As the historical frame weight, As the weight of the current frame, the previous frame is mixed into the current frame using a weighted mixing method.

[0064] Among them, due to the high detail and low noise components Contains an estimate of the opacity of the current pixel through the volume data, so in some embodiments, high detail low noise components Denoising also includes performing spatiotemporal denoising on the opacity of the pixels in the current frame, which can effectively distinguish between details and noise in the image.

[0065] Step S330: Using the denoised high-detail low-noise component as an auxiliary feature to guide spatial denoising of the low-detail high-noise component.

[0066] In some embodiments, for low detail high noise components Perform time series denoising to smooth out the noise in the time dimension. Specifically: high-detail low-noise components after time and space denoising The depth value is used as an auxiliary guiding feature, and the filter kernel is expanded to increase the coverage of the filter, so as to perform denoising in a larger area. For example, the filter kernel can be expanded three times to increase the filtering range.

[0067] In some embodiments, step S200, performing shading estimation on the volume data based on the Monte Carlo algorithm and generating radiation samples, further includes applying a hybrid acceleration structure to sample the volume data, specifically: The three-dimensional space of the volume data is divided into a plurality of macro units, non-empty macro units are determined, and an octree is constructed to recursively divide the macro units into a plurality of octree nodes; the octree nodes are linearly stored in spatial order, wherein each node contains an index of a parent node, indexes of all child nodes, and a range of values ​​of voxels in a spatial region represented by most octree nodes.

[0068] In the embodiment of the present application, compared with conventional accelerators (such as octrees and VDBs), the hybrid acceleration structure based on macro units and octrees can combine the advantages of octrees in skipping empty areas and the fast traversal of the macro unit's 3D-DDA (three-dimensional non-continuous deformation analysis) sampling technology in the volume space.

[0069] For example, in combination Figure 5 As shown, Figure 5 The spatial partition diagram of the macrogrid and octree hybrid acceleration structure of a real-time medical image 3D volume rendering method with full dynamic illumination provided by the present application is shown. Each partition block represents an octree node, and for changes in the transfer function (mapping voxel values ​​to color and opacity), the maximum extinction coefficient of each node can be updated according to the voxel value range stored in the node. . Indicates the degree to which light is absorbed in voxels during volume path tracing sampling. If the value in the node is zero, that is, there are not enough voxel values ​​in the node to affect the propagation of light, the sampling and calculation of the node can be skipped, thereby greatly improving rendering efficiency.

[0070] In some embodiments, the present application also provides a computer program product, which includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the above-mentioned real-time medical image three-dimensional rendering method with full dynamic lighting is implemented.

[0071] It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems or computer program products. Those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0072] The above description is only an implementation mode of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, should be included in the protection scope of the present invention.

Claims

1. A method for real-time medical image 3D volume rendering with full dynamic illumination, characterized in that: include: Acquire volume data, where the volume data is three-dimensional information corresponding to the three-dimensional object; Performing coloring estimation on the volume data based on a Monte Carlo algorithm and generating radiation samples; Performing feature decoupling on the radiation sample to decompose it into a low-detail high-noise component and a high-detail low-noise component, wherein the low-detail high-noise component contains information related to light source radiation and light transmittance, and the high-detail low-noise component contains scattering properties related to surface material; De-noising the low-detail high-noise component and the high-detail low-noise component respectively; The denoised low detail high noise component and the high detail low noise component are combined to generate a denoised image.

2. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 1, characterized in that: The method of performing coloring estimation on the volume data based on the Monte Carlo algorithm and generating radiation samples includes: Pre-constructing a surface-medium coexistence shading model, wherein the surface-medium coexistence shading model defines a composite particle having both body scattering characteristics and surface scattering characteristics; The scattering distribution function of the composite particle is defined as: in, represents the BRDF value, represents the volume scattering probability at the sampling point x, represents the surface scattering probability at the sampling point x, represents the phase function.

3. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 2, characterized in that: The method of performing coloring estimation on the volume data based on the Monte Carlo algorithm and generating radiation samples further includes: The direct illumination estimation of the volume data is performed based on the surface medium coexistence shading model, which is expressed as: in, ) represents the sampling point x to the sampling point The light transmittance between Represents the sampling point x The light intensity in a direction, Indicates that the light source is at the sampling point The sampling density of .

4. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 1, characterized in that: The radiation sample is feature decoupled to decompose it into a low-detail high-noise component and a high-detail low-noise component, including: Decomposing the radiation sample into the low detail high noise components and the high detail low noise components , recorded as: in, represents the radiation of the light source, represents the refractive index of light, Indicates that the light source is at the sampling point The sampling density of Represents the lighting estimation result.

5. The method for real-time medical image 3D volume rendering with full dynamic illumination according to any one of claims 1 to 4, characterized in that: The denoising of the low-detail high-noise component and the high-detail low-noise component respectively further includes: Determine the first real scattering event, and when the particle density of the sampling point of the real scattering event exceeds a preset threshold, record the depth value of the corresponding sampling point; Selecting a minimum depth value from the recorded depth values ​​of corresponding sampling points, and using the minimum depth value for spatial denoising of the high-detail low-noise component and / or spatial denoising of the low-detail high-noise component; When the particle density of all sampling points is lower than the preset threshold, the middle value of the depth values ​​of all sampling points is selected, and the middle depth value is used for spatial denoising of the high-detail low-noise component and / or spatial denoising of the low-detail high-noise component.

6. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 4, characterized in that: The denoising of the low-detail high-noise component and the high-detail low-noise component respectively comprises: High detail low noise components Perform spatial denoising; High detail low noise components Performing time series denoising, wherein the time series denoising includes: Calculate the distance of the geometric auxiliary features of the relevant pixels ; The distance of the geometric auxiliary feature of the relevant pixel in exponential form As the historical frame weight; The current frame weight is recorded as: Blend the previous frame rendering result into the current frame.

7. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 6, characterized in that: High detail low noise components Denoising also includes performing spatiotemporal denoising on the opacity of the pixels in the current frame.

8. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 6, characterized in that: The denoising of the low-detail high-noise component and the high-detail low-noise component respectively comprises: The denoised high-detail low-noise component is used as an auxiliary feature to guide spatial denoising of the low-detail high-noise component.

9. The method for real-time medical image 3D volume rendering with full dynamic illumination according to claim 1, characterized in that: The method of performing coloring estimation on the volume data based on the Monte Carlo algorithm and generating radiation samples further includes: The volume data is sampled using a hybrid acceleration structure, specifically: Dividing the three-dimensional space of the volume data into a plurality of macro units; Determining a non-empty macrocell and constructing an octree to recursively divide the macrocell into a plurality of octree nodes; The octree nodes are linearly stored in spatial order, wherein each node contains an index of a parent node, indexes of all child nodes, and a range of values ​​of voxels within a spatial region represented by most octree nodes.

10. A computer program product, comprising a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the method for real-time medical image three-dimensional volume rendering with full dynamic illumination according to any one of claims 1 to 9 is implemented.

Images