Denoising method, system and equipment based on multi-model adaptive mask and medium

Abstract

The invention relates to an image denoising technology, and discloses a denoising method based on a multi-model adaptive mask, which comprises the following steps: carrying out mandible segmentation and tooth segmentation on image data subjected to format conversion processing to obtain a mandible segmentation image and a tooth segmentation image, carrying out detail segmentation and foreign matter segmentation on the format conversion image to obtain a tooth segmentation image; the lower jaw segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image are collected to obtain segmentation result data, an adaptive weight is generated according to the segmentation result data, and noise reduction processing is performed on the segmentation result data to obtain a noise reduction segmentation result. And carrying out weighted superposition fusion on the noise reduction segmentation result according to the adaptive weight to obtain a final denoised image. The invention further provides a denoising device and equipment based on the multi-model adaptive mask and a medium. According to the invention, the image denoising effect can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image denoising, and in particular to a denoising method, system, device and medium based on multi-model adaptive masking. Background Art

[0002] In the field of dental image processing, accurate denoising and enhancement are crucial to diagnostic results, especially when processing images involving structures such as teeth, implants, and jaws, where different structures have different requirements for noise and enhancement.

[0003] Existing single denoising models usually process the image as a whole, but fail to distinguish specific anatomical structures. This approach often results in insufficient or excessive noise removal in certain areas, resulting in serious loss of details. In addition, the representation of teeth, details, implants, and mandibles in images varies greatly. Implants are easily disturbed by artifacts and strong noise due to their material properties; teeth have rich texture details and require high detail retention; mandibles have special requirements for enhancement and noise reduction due to differences in density and morphology, and detailed areas are mixed with noise and are difficult to distinguish. A unified processing method cannot meet these different needs. Summary of the invention

[0004] The present invention provides a denoising method, system, device and medium based on multi-model adaptive masking, the main purpose of which is to solve the problem of poor effect of existing denoising methods and achieve adaptation to different denoising requirements of different parts of the oral cavity.

[0005] To achieve the above object, the present invention provides a denoising method based on multi-model adaptive masking, comprising:

[0006] Performing format conversion processing on the pre-acquired image data to obtain a format-converted image;

[0007] performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performing teeth segmentation according to the format conversion image to obtain a teeth segmentation image;

[0008] Using a pre-trained first residual neural network model to perform detail segmentation on the image according to the format conversion to obtain a segmentation feature image;

[0009] Using a pre-trained rated second residual neural network model to perform foreign body segmentation according to the format conversion image, to obtain a foreign body segmentation image;

[0010] Summarizing the mandibular segmentation image, the teeth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data;

[0011] The segmentation result data is subjected to denoising processing to obtain a denoised segmentation result, and the denoised segmentation result is subjected to weighted superposition fusion according to the adaptive weight to obtain a final denoised image.

[0012] Optionally, performing format conversion processing on the pre-acquired image data to obtain a format-converted image includes:

[0013] Acquire a pixel value range of the image data;

[0014] Normalizing the image data according to the pixel value range to obtain normalized data;

[0015] The normalized data is converted into a 32-bit floating point number to obtain a format-converted image.

[0016] Optionally, performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image includes:

[0017] Using a first dense link unit in a pre-trained first dense link model to perform four convolution operations on the format conversion image to obtain a first convolution feature, and splicing the first convolution feature with the format conversion image based on a dimensional channel to obtain a first splicing feature;

[0018] Using a second dense link unit in the first dense link model to perform four convolution operations on the first splicing feature to obtain a second convolution feature, and splicing the format conversion image, the first splicing feature, and the second convolution feature based on a dimensional channel to obtain a second splicing feature;

[0019] Using the third dense link unit in the first dense link model to perform four convolution operations on the second splicing feature to obtain a third convolution feature, and splicing the format conversion image, the first splicing feature, the second convolution feature and the third convolution feature based on a dimensional channel to obtain a third splicing feature;

[0020] Using a fourth dense link unit in the first dense link model to perform four convolution operations on the third concatenation feature to obtain a fourth convolution feature, and concatenating the format conversion image, the first concatenation feature, the second convolution feature, the third convolution feature and the fourth convolution feature based on a dimensional channel to obtain a fourth concatenation feature;

[0021] The fourth convolution operation is performed four times on the fourth splicing feature using the fourth dense link unit in the first dense link model to obtain a fifth convolution feature, and the format conversion image, the first splicing feature, the second convolution feature, the third convolution feature, the fourth convolution feature and the fifth convolution feature are spliced ​​based on the dimensional channel to obtain a jaw segmentation image.

[0022] Optionally, the using a pre-trained first residual neural network model to perform detail segmentation according to the format conversion image includes:

[0023] Using a first residual unit in the first residual neural network model to perform a convolution operation on the format conversion image to obtain a first convolution result, and adding the format conversion image to the first convolution result to obtain a first addition result;

[0024] Using a second residual unit in the first residual neural network model to perform a convolution operation on the first convolution result to obtain a second convolution result, and adding the second convolution result, the first convolution result, and the first addition result to obtain a second addition result;

[0025] Using the third residual unit in the first residual neural network model to perform a convolution operation on the second convolution result to obtain a third convolution result, and adding the third convolution result, the second convolution result, the first addition result, and the second addition result to obtain a third addition result;

[0026] Using a fourth residual unit in the first residual neural network model to perform a convolution operation on the third convolution result to obtain a fourth convolution result, and adding the fourth convolution result, the third convolution result, the first addition result, the second addition result, and the third addition result to obtain a fourth addition result;

[0027] The fourth convolution result is convolved by the fifth residual unit in the first residual neural network model to obtain a fifth convolution result, and the fifth convolution result, the fourth convolution result, the first addition result, the second addition result, the third addition result, and the fourth addition result are added to obtain a segmentation feature image.

[0028] Optionally, generating an adaptive weight according to the segmentation result data includes:

[0029] Generate region mask data for each segmented region according to the segmentation result data;

[0030] Performing a preset number of corrosion operations on the regional mask data to obtain a corrosion operation result;

[0031] Performing a preset number of expansion operations on the region mask data to obtain an expansion operation result;

[0032] The erosion operation results and the dilation operation results of each segmented region are summed pixel by pixel and normalized to obtain the adaptive weight of each segmented region.

[0033] Optionally, performing denoising on the segmentation result data to obtain a denoised segmentation result includes:

[0034] Calculating the noise intensity of each segmented area in the segmentation result data, and setting a smoothness control parameter according to the noise intensity;

[0035] Calculating a search window size according to the noise intensity, and calculating a similarity window size according to the noise intensity;

[0036] The segmentation result data is subjected to denoising processing by using a non-local means algorithm according to the search window size, the similarity window size, and the smoothness control parameter to obtain a denoised segmentation result.

[0037] Optionally, calculating the search window size according to the noise intensity includes:

[0038] The search window size is calculated using the following formula:

[0039]

[0040] Wherein, w is the search window size, w max is the preset maximum window parameter, w min is the preset minimum window parameter, S is the noise intensity of the current segmented area in the noise intensity, S max is the signal-to-noise ratio in the area with the maximum noise intensity, S min is the signal-to-noise in the area of ​​minimum noise intensity.

[0041] In order to solve the above problems, the present invention also provides a denoising device based on multi-model adaptive masking, the device comprising:

[0042] A data conversion module is used to perform format conversion processing on the pre-acquired image data to obtain a format-converted image;

[0043] A first segmentation module is used to perform jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and to perform teeth segmentation according to the format conversion image to obtain a teeth segmentation image;

[0044] A second segmentation module is used to perform detail segmentation according to the format conversion image using a pre-trained first residual neural network model to obtain a segmentation feature image, and to perform foreign body segmentation according to the format conversion image using a pre-trained second residual neural network model to obtain a foreign body segmentation image;

[0045] A weight generation module, used for summarizing the mandibular segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data;

[0046] The fusion denoising module is used to perform denoising on the segmentation result data to obtain a denoised segmentation result, and perform weighted superposition fusion on the denoised segmentation result according to the adaptive weight to obtain a final denoised image.

[0047] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0048] at least one processor;

[0049] and, a memory communicatively coupled to the at least one processor;

[0050] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the above-mentioned denoising method based on multi-model adaptive mask.

[0051] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned denoising method based on multi-model adaptive masking.

[0052] The embodiment of the present invention performs format conversion processing on the pre-acquired image data to obtain a format conversion image, performs jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performs teeth segmentation according to the format conversion image to obtain a teeth segmentation image, performs detail segmentation according to the format conversion image using a pre-trained first residual neural network model to obtain a segmentation feature image, performs foreign body segmentation according to the format conversion image using a pre-trained second residual neural network model to obtain a foreign body segmentation image, summarizes the jaw segmentation image, the teeth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, generates adaptive weights according to the segmentation result data, performs denoising processing on the segmentation result data to obtain a denoised segmentation result, and performs weighted superposition fusion on the denoised segmentation result according to the adaptive weight to obtain a final denoised image. Therefore, the denoising method, device, electronic device and computer-readable storage medium based on multi-model adaptive mask proposed in the present invention can solve the problem of poor effect of existing denoising methods and achieve adaptation to different denoising requirements of different parts of the oral cavity. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a process flow of a denoising method based on multi-model adaptive masking provided by an embodiment of the present invention;

[0054] Figure 2 A functional module diagram of a denoising device based on multi-model adaptive masking provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of the structure of an electronic device for implementing the denoising method based on multi-model adaptive masking provided by an embodiment of the present invention;

[0056] Figure 4 A schematic flow chart of a multi-model adaptive mask denoising model is provided for one embodiment of the present invention.

[0057] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0058] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0059] The embodiment of the present application provides a denoising method based on multi-model adaptive masking. The execution subject of the denoising method based on multi-model adaptive masking includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the denoising method based on multi-model adaptive masking can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0060] Reference Figure 1 FIG. 1 is a flow chart of a denoising method based on multi-model adaptive masking according to an embodiment of the present invention. In this embodiment, the denoising method based on multi-model adaptive masking includes:

[0061] S1. Perform format conversion processing on pre-acquired image data to obtain a format converted image.

[0062] In the embodiment of the present invention, the pre-acquired image data may be a medical image, such as an oral medical image.

[0063] In an embodiment of the present invention, the pre-acquired image data is format-converted to obtain a format-converted image, which means converting the image data into a 32-bit floating-point data format, for example, converting the pixel value of each pixel of the image data into floating-point data.

[0064] In detail, by converting the image data into a 32-bit floating point data format, the dynamic range of the data can be improved, higher data accuracy can be provided for subsequent processing, and integer overflow can be avoided in subsequent processing.

[0065] In the embodiment of the present invention, the step of performing format conversion processing on the pre-acquired image data to obtain a format converted image includes:

[0066] Acquire a pixel value range of the image data;

[0067] Normalizing the image data according to the pixel value range to obtain normalized data;

[0068] The normalized data is converted into a 32-bit floating point number to obtain a format-converted image.

[0069] In the embodiment of the present invention, a format conversion process is performed on pre-acquired image data to obtain a format conversion image, thereby improving the accuracy of subsequent segmentation of the format conversion image.

[0070] S2. Performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performing teeth segmentation according to the format conversion image to obtain a teeth segmentation image.

[0071] In the embodiment of the present invention, the jaw segmentation is performed according to the format conversion image to obtain the jaw segmentation image, which is performed according to the format conversion image using a pre-trained first dense link model.

[0072] In detail, the first densely linked model is a deep learning network architecture whose core feature is dense connection, that is, each layer in the network is directly connected to all previous layers. This design enables each layer to access the feature maps of all previous layers, thereby achieving feature reuse.

[0073] In detail, the first dense link model includes 5 different dense link units, each dense link unit includes four convolution unit layers, and each convolution unit layer includes a 3×3 convolution layer, an activation layer, and a batch normalization layer.

[0074] In the embodiment of the present invention, the step of performing jaw segmentation according to the format conversion image to obtain the jaw segmentation image includes:

[0075] Using a first dense link unit in a pre-trained first dense link model to perform four convolution operations on the format conversion image to obtain a first convolution feature, and splicing the first convolution feature with the format conversion image based on a dimensional channel to obtain a first splicing feature;

[0076] Using a second dense link unit in the first dense link model to perform four convolution operations on the first splicing feature to obtain a second convolution feature, and splicing the format conversion image, the first splicing feature, and the second convolution feature based on a dimensional channel to obtain a second splicing feature;

[0077] Using the third dense link unit in the first dense link model to perform four convolution operations on the second splicing feature to obtain a third convolution feature, and splicing the format conversion image, the first splicing feature, the second convolution feature and the third convolution feature based on a dimensional channel to obtain a third splicing feature;

[0078] Using a fourth dense link unit in the first dense link model to perform four convolution operations on the third concatenation feature to obtain a fourth convolution feature, and concatenating the format conversion image, the first concatenation feature, the second convolution feature, the third convolution feature and the fourth convolution feature based on a dimensional channel to obtain a fourth concatenation feature;

[0079] The fourth convolution operation is performed four times on the fourth splicing feature using the fourth dense link unit in the first dense link model to obtain a fifth convolution feature, and the format conversion image, the first splicing feature, the second convolution feature, the third convolution feature, the fourth convolution feature and the fifth convolution feature are spliced ​​based on the dimensional channel to obtain a jaw segmentation image.

[0080] In detail, performing four convolution operations on the format-converted image using the first dense link unit in the pre-trained first dense link model refers to performing convolution operations in sequence using four convolution units in the first dense link unit.

[0081] In detail, the step of splicing the format-converted image, the first splicing feature, and the second convolution feature based on the dimensional channel to obtain the second splicing feature refers to splicing the first splicing feature and the second convolution feature in the second dimension.

[0082] In the embodiment of the present invention, the teeth segmentation is performed according to the format conversion image to obtain the teeth segmentation image, which is performed according to the format conversion image using a pre-trained second dense link model.

[0083] In detail, the tooth segmentation is performed according to the format conversion image to obtain the tooth segmentation image, which is achieved by sequentially performing convolution operations on multiple dense link units in the second dense link model, and each dense link unit splices the result of the convolution operation with the output result of each previous dense link unit until the last dense link unit completes the splicing to obtain the tooth segmentation result.

[0084] In an embodiment of the present invention, the second dense link model includes 5 different dense link units, each dense link unit includes four convolution unit layers, and each convolution unit layer includes a 3x3 convolution layer, an activation layer and a batch normalization layer.

[0085] In the embodiment of the present invention, the mandible is segmented according to the format conversion image to obtain a mandible segmentation image, and the teeth are segmented according to the format conversion image to obtain a teeth segmentation image, thereby improving the efficiency of subsequent aggregation to obtain segmentation result data.

[0086] S3. Use the pre-trained first residual neural network model to perform detail segmentation on the image according to the format conversion to obtain a segmentation feature image.

[0087] In an embodiment of the present invention, the first residual neural network model is a deep convolutional neural network (CNN) architecture, which solves the degradation problem in deep network training by introducing "residual learning". The core idea of ​​the residual learning is to let the network learn the residual (i.e., difference) between the input and output, rather than directly learning the mapping function. Residual learning is achieved by introducing shortcut connections. These connections allow the input of earlier layers in the network to be directly passed to the later layers, bypassing some intermediate layers, so that the gradient can be directly passed to the front layer, effectively alleviating the gradient vanishing problem, so that deeper networks can be trained.

[0088] In an embodiment of the present invention, the first residual neural network model includes 5 different residual link units, the residual link unit includes two convolution units and a 1x1 convolution layer, and each convolution unit layer includes a 3x3 convolution layer, an activation layer, and a batch normalization layer.

[0089] In the embodiment of the present invention, the use of the pre-trained first residual neural network model to perform detail segmentation according to the format conversion image refers to segmenting the trabeculae, bilateral bone joints and other parts in the format conversion image.

[0090] In an embodiment of the present invention, the step of using a pre-trained first residual neural network model to perform detail segmentation according to the format conversion image includes:

[0091] Using a first residual unit in the first residual neural network model to perform a convolution operation on the format conversion image to obtain a first convolution result, and adding the format conversion image to the first convolution result to obtain a first addition result;

[0092] Using a second residual unit in the first residual neural network model to perform a convolution operation on the first convolution result to obtain a second convolution result, and adding the second convolution result, the first convolution result, and the first addition result to obtain a second addition result;

[0093] Using the third residual unit in the first residual neural network model to perform a convolution operation on the second convolution result to obtain a third convolution result, and adding the third convolution result, the second convolution result, the first addition result, and the second addition result to obtain a third addition result;

[0094] Using a fourth residual unit in the first residual neural network model to perform a convolution operation on the third convolution result to obtain a fourth convolution result, and adding the fourth convolution result, the third convolution result, the first addition result, the second addition result, and the third addition result to obtain a fourth addition result;

[0095] The fourth convolution result is convolved by the fifth residual unit in the first residual neural network model to obtain a fifth convolution result, and the fifth convolution result, the fourth convolution result, the first addition result, the second addition result, the third addition result, and the fourth addition result are added to obtain a segmentation feature image.

[0096] In the embodiment of the present invention, by using a pre-trained first residual neural network model to convert the image according to the format, detail segmentation is performed to obtain a segmentation feature image, thereby improving the efficiency of subsequent aggregation to obtain segmentation result data.

[0097] S4. Using the pre-trained rated second residual neural network model, the image is converted according to the format to perform foreign body segmentation to obtain a foreign body segmentation image.

[0098] S5. Summarize the mandibular segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generate adaptive weights according to the segmentation result data.

[0099] In the embodiment of the present invention, generating an adaptive weight according to the segmentation result data includes:

[0100] Generate region mask data for each segmented region according to the segmentation result data;

[0101] Performing a preset number of corrosion operations on the regional mask data to obtain a corrosion operation result;

[0102] Performing a preset number of expansion operations on the region mask data to obtain an expansion operation result;

[0103] The erosion operation results and the dilation operation results of each segmented region are summed pixel by pixel and normalized to obtain the adaptive weight of each segmented region.

[0104] In detail, performing a preset number of corrosion operations on the region mask data refers to performing five convolution operations on the region mask data using a convolution kernel with a size of 3*3 and minimum values ​​for all elements.

[0105] Specifically, performing a preset number of expansion operations on the region mask data refers to performing five convolution operations on the region mask data using a convolution kernel of size 3*3 and with all elements taking the maximum value.

[0106] In an embodiment of the present invention, segmentation result data is obtained by aggregating the mandibular segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image, and adaptive weights are generated based on the segmentation result data, thereby improving the accuracy of subsequent weighted superposition fusion and improving the denoising effect.

[0107] S6. Perform denoising on the segmentation result data to obtain a denoised segmentation result, and perform weighted superposition fusion on the denoised segmentation result according to the adaptive weight to obtain a final denoised image.

[0108] In the embodiment of the present invention, the step of performing noise reduction processing on the segmentation result data to obtain the noise reduction segmentation result includes:

[0109] Calculating the noise intensity of each segmented area in the segmentation result data, and setting a smoothness control parameter according to the noise intensity;

[0110] Calculating a search window size according to the noise intensity, and calculating a similarity window size according to the noise intensity;

[0111] The segmentation result data is subjected to denoising processing by using a non-local means algorithm according to the search window size, the similarity window size, and the smoothness control parameter to obtain a denoised segmentation result.

[0112] In detail, calculating the search window size according to the noise intensity includes:

[0113] The search window size is calculated using the following formula:

[0114]

[0115] Wherein, w is the search window size, w max is the preset maximum window parameter, w min is the preset minimum window parameter, S is the noise intensity of the current segmented area in the noise intensity, S max is the signal-to-noise ratio in the area with the maximum noise intensity, S min is the signal-to-noise in the area of ​​minimum noise intensity.

[0116] In detail, calculating the noise intensity of each segmented area in the segmentation result data and setting the smoothness control parameter according to the noise intensity means setting the smoothing control parameter of the area with the largest noise intensity to 1 and setting the smoothing control parameter of the area with the smallest noise intensity to 0.5.

[0117] In detail, the similarity window size is calculated according to the noise intensity, and is calculated by the same method as the search window size sampling is calculated according to the noise intensity, which will not be repeated here.

[0118] In detail, the preset maximum window parameter can be 25, and the preset minimum window parameter can be 5.

[0119] In detail, the non-local means algorithm is an image denoising algorithm, and its core idea is to use redundant information in the image for denoising. It assumes that there is similarity between the pixels in the image and the pixels in its non-local area. For example, in a natural landscape image, the pixels in the sky may have similar grayscale values ​​at different locations. For each pixel in the image, the NLM algorithm will search for similar pixel areas in the entire image, and then perform weighted averaging on the pixel based on these similar blocks, so as to achieve the purpose of denoising.

[0120] In the embodiment of the present invention, the weighted superposition fusion is an image processing technology that generates a new image by combining information of multiple images according to certain weights, so that the new image can contain all key information and features of the original images.

[0121] In the embodiment of the present invention, by performing denoising processing on the segmentation result data to obtain a denoised segmentation result, the denoised segmentation result is weightedly superimposed and fused according to the adaptive weight to obtain a final denoised image, thereby improving the image denoising effect.

[0122] Reference Figure 4As shown, a flow chart of a multi-model adaptive mask denoising model is provided for an embodiment of the present invention, wherein M1, M2, M3, M4, and M5 are five different densely linked units of the first densely linked model; D1, D2, D3, D4, and D5 are five different densely linked units of the second densely linked model; O1, O2, O3, O4, and O5 are five different residual units of the first residual neural network model; and T1, T2, T3, T4, and T5 are five different residual units of the second residual neural network model.

[0123] like Figure 2 , which is a functional module diagram of a denoising device based on multi-model adaptive masking provided by an embodiment of the present invention.

[0124] The denoising device 100 based on multi-model adaptive mask of the present invention can be installed in an electronic device. According to the functions implemented, the denoising device 100 based on multi-model adaptive mask can include a data conversion module 101, a first segmentation module 102, a second segmentation module 103, a weight generation module 104 and a fusion denoising module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, which are stored in the memory of the electronic device.

[0125] In this embodiment, the functions of each module / unit are as follows:

[0126] The data conversion module 101 is used to perform format conversion processing on the pre-acquired image data to obtain a format-converted image;

[0127] The first segmentation module 102 is used to perform jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and to perform teeth segmentation according to the format conversion image to obtain a teeth segmentation image;

[0128] The second segmentation module 103 is used to perform detail segmentation according to the format conversion image using a pre-trained first residual neural network model to obtain a segmentation feature image, and to perform foreign body segmentation according to the format conversion image using a pre-trained second residual neural network model to obtain a foreign body segmentation image;

[0129] The weight generation module 104 is used to aggregate the mandibular segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generate adaptive weights according to the segmentation result data;

[0130] The fusion denoising module 105 is used to perform denoising on the segmentation result data to obtain a denoised segmentation result, and perform weighted superposition fusion on the denoised segmentation result according to the adaptive weight to obtain a final denoised image.

[0131] In detail, each module described in the denoising device 100 based on multi-model adaptive masking in the embodiment of the present invention is used in the same manner as described above. Figures 1 to 3 The denoising method based on multi-model adaptive masking described in the invention has the same technical means and can produce the same technical effect, so it will not be repeated here.

[0132] like Figure 3 , which is a schematic diagram of the structure of an electronic device of a denoising method based on multi-model adaptive masking provided by an embodiment of the present invention.

[0133] The electronic device 1 may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a denoising program based on multi-model adaptive masks.

[0134] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a denoising program based on multi-model adaptive masks, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device and process data.

[0135] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 may also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 11 may also include both an internal storage unit of the electronic device and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device, such as the code of a denoising program based on a multi-model adaptive mask, but also can be used to temporarily store data that has been output or is to be output.

[0136] The communication bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0137] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in this embodiment, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visual user interface.

[0138] The figure only shows an electronic device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0139] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0140] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0141] The denoising program based on multi-model adaptive mask stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0142] Performing format conversion processing on the pre-acquired image data to obtain a format-converted image;

[0143] performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performing teeth segmentation according to the format conversion image to obtain a teeth segmentation image;

[0144] Using a pre-trained first residual neural network model to perform detail segmentation on the image according to the format conversion to obtain a segmentation feature image;

[0145] Using a pre-trained rated second residual neural network model to perform foreign body segmentation according to the format conversion image, to obtain a foreign body segmentation image;

[0146] Summarizing the mandibular segmentation image, the teeth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data;

[0147] The segmentation result data is subjected to denoising processing to obtain a denoised segmentation result, and the denoised segmentation result is subjected to weighted superposition fusion according to the adaptive weight to obtain a final denoised image.

[0148] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, which will not be repeated here.

[0149] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0150] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0151] Performing format conversion processing on the pre-acquired image data to obtain a format-converted image;

[0152] performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performing teeth segmentation according to the format conversion image to obtain a teeth segmentation image;

[0153] Using a pre-trained first residual neural network model to perform detail segmentation on the image according to the format conversion to obtain a segmentation feature image;

[0154] Using a pre-trained rated second residual neural network model to perform foreign body segmentation according to the format conversion image, to obtain a foreign body segmentation image;

[0155] Summarizing the mandibular segmentation image, the teeth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data;

[0156] The segmentation result data is subjected to denoising processing to obtain a denoised segmentation result, and the denoised segmentation result is subjected to weighted superposition fusion according to the adaptive weight to obtain a final denoised image.

[0157] In the embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0158] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0159] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0160] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0161] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0162] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0163] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A denoising method based on multi-model adaptive masking, characterized in that: The method comprises: Performing format conversion processing on the pre-acquired image data to obtain a format converted image; performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and performing teeth segmentation according to the format conversion image to obtain a teeth segmentation image; Using a pre-trained first residual neural network model to perform detail segmentation on the image according to the format conversion to obtain a segmentation feature image; Using a pre-trained rated second residual neural network model to perform foreign body segmentation according to the format conversion image, to obtain a foreign body segmentation image; Summarizing the mandibular segmentation image, the teeth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data; The segmentation result data is subjected to denoising processing to obtain a denoised segmentation result, and the denoised segmentation result is subjected to weighted superposition fusion according to the adaptive weight to obtain a final denoised image.

2. The denoising method based on multi-model adaptive masking as claimed in claim 1, characterized in that: The step of performing format conversion processing on the pre-acquired image data to obtain a format converted image includes: Acquire a pixel value range of the image data; Normalizing the image data according to the pixel value range to obtain normalized data; The normalized data is converted into a 32-bit floating point number to obtain a format-converted image.

3. The denoising method based on multi-model adaptive masking as claimed in claim 2, characterized in that: The step of performing jaw segmentation according to the format conversion image to obtain a jaw segmentation image includes: Using a first dense link unit in a pre-trained first dense link model to perform four convolution operations on the format conversion image to obtain a first convolution feature, and splicing the first convolution feature with the format conversion image based on a dimensional channel to obtain a first splicing feature; Using a second dense link unit in the first dense link model to perform four convolution operations on the first splicing feature to obtain a second convolution feature, and splicing the format conversion image, the first splicing feature, and the second convolution feature based on a dimensional channel to obtain a second splicing feature; Using the third dense link unit in the first dense link model to perform four convolution operations on the second splicing feature to obtain a third convolution feature, and splicing the format conversion image, the first splicing feature, the second convolution feature and the third convolution feature based on a dimensional channel to obtain a third splicing feature; Using a fourth dense link unit in the first dense link model to perform four convolution operations on the third concatenation feature to obtain a fourth convolution feature, and concatenating the format conversion image, the first concatenation feature, the second convolution feature, the third convolution feature and the fourth convolution feature based on a dimensional channel to obtain a fourth concatenation feature; The fourth convolution operation is performed four times on the fourth splicing feature using the fourth dense link unit in the first dense link model to obtain a fifth convolution feature, and the format conversion image, the first splicing feature, the second convolution feature, the third convolution feature, the fourth convolution feature and the fifth convolution feature are spliced ​​based on the dimensional channel to obtain a jaw segmentation image.

4. The denoising method based on multi-model adaptive masking as claimed in claim 1, characterized in that: The method of using the pre-trained first residual neural network model to convert the image into a detailed segmentation format according to the format includes: Using a first residual unit in the first residual neural network model to perform a convolution operation on the format conversion image to obtain a first convolution result, and adding the format conversion image to the first convolution result to obtain a first addition result; Using a second residual unit in the first residual neural network model to perform a convolution operation on the first convolution result to obtain a second convolution result, and adding the second convolution result, the first convolution result, and the first addition result to obtain a second addition result; Using the third residual unit in the first residual neural network model to perform a convolution operation on the second convolution result to obtain a third convolution result, and adding the third convolution result, the second convolution result, the first addition result, and the second addition result to obtain a third addition result; Using a fourth residual unit in the first residual neural network model to perform a convolution operation on the third convolution result to obtain a fourth convolution result, and adding the fourth convolution result, the third convolution result, the first addition result, the second addition result, and the third addition result to obtain a fourth addition result; The fourth convolution result is convolved by the fifth residual unit in the first residual neural network model to obtain a fifth convolution result, and the fifth convolution result, the fourth convolution result, the first addition result, the second addition result, the third addition result, and the fourth addition result are added to obtain a segmentation feature image.

5. The denoising method based on multi-model adaptive masking as claimed in claim 1, characterized in that: The step of generating an adaptive weight according to the segmentation result data comprises: Generate region mask data for each segmented region according to the segmentation result data; Performing a preset number of corrosion operations on the regional mask data to obtain a corrosion operation result; Performing a preset number of expansion operations on the region mask data to obtain an expansion operation result; The erosion operation results and the dilation operation results of each segmented region are summed pixel by pixel and normalized to obtain the adaptive weight of each segmented region.

6. The denoising method based on multi-model adaptive masking as claimed in claim 1, characterized in that: The performing denoising on the segmentation result data to obtain a denoised segmentation result includes: Calculating the noise intensity of each segmented area in the segmentation result data, and setting a smoothness control parameter according to the noise intensity; Calculating a search window size according to the noise intensity, and calculating a similarity window size according to the noise intensity; The segmentation result data is subjected to denoising processing by using a non-local means algorithm according to the search window size, the similarity window size, and the smoothness control parameter to obtain a denoised segmentation result.

7. The denoising method based on multi-model adaptive masking as claimed in claim 6, characterized in that: The calculating the search window size according to the noise intensity comprises: The search window size is calculated using the following formula: Wherein, w is the search window size, w max is the preset maximum window parameter, w min is the preset minimum window parameter, S is the noise intensity of the current segmented area in the noise intensity, S max is the signal-to-noise ratio in the area with the maximum noise intensity, S min is the signal-to-noise in the area of ​​minimum noise intensity.

8. A denoising device based on multi-model adaptive masking, characterized in that: The device comprises: A data conversion module is used to perform format conversion processing on the pre-acquired image data to obtain a format-converted image; A first segmentation module is used to perform jaw segmentation according to the format conversion image to obtain a jaw segmentation image, and to perform teeth segmentation according to the format conversion image to obtain a teeth segmentation image; A second segmentation module is used to perform detail segmentation according to the format conversion image using a pre-trained first residual neural network model to obtain a segmentation feature image, and to perform foreign body segmentation according to the format conversion image using a pre-trained second residual neural network model to obtain a foreign body segmentation image; A weight generation module, used for summarizing the mandibular segmentation image, the tooth segmentation image, the segmentation feature image and the foreign body segmentation image to obtain segmentation result data, and generating an adaptive weight according to the segmentation result data; The fusion denoising module is used to perform denoising on the segmentation result data to obtain a denoised segmentation result, and perform weighted superposition fusion on the denoised segmentation result according to the adaptive weight to obtain a final denoised image.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; Wherein, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the denoising method based on multi-model adaptive mask as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the denoising method based on multi-model adaptive masking as described in any one of claims 1 to 7 is implemented.

Images