Image denoising method and device

Abstract

The present invention provides a method and device for image denoising, characterized in that a deep learning method is used to train to obtain relevant information required for image denoising, so as to achieve image denoising, and a convolutional neural network can be used to obtain relevant information for image denoising to achieve final denoising. The present invention can use different denoising methods for images of different scales to improve the denoising effect.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing, and in particular to an image denoising method and device. Background Art

[0002] Image denoising has always been a basic work in the field of image processing and computer vision. In recent years, deep learning (DN) has been widely used in the field of image denoising. The deep learning method is used to train and obtain the relevant information required for image denoising to achieve image denoising. For example, convolutional neural networks (CNN) can be used to obtain relevant information for image denoising to achieve final denoising. For images of different scales, it is necessary to use different denoising methods to improve the denoising effect. Summary of the invention

[0003] Based on this, it is necessary to provide an image denoising method and device to address the above technical issues.

[0004] According to one aspect of the present disclosure, a data denoising method is provided, characterized in that the method comprises:

[0005] receiving noisy image data;

[0006] Calculating bias vector data and scale selection data of the image data using a bias network model;

[0007] For noisy image data at different scales, the filter weights and the bias vector data are used to calculate the denoised image output;

[0008] According to the scale selection data, the denoised images calculated at different scales are fused into a final denoised image for output.

[0009] According to another aspect of the present disclosure, a data denoising device is provided, characterized in that the device comprises:

[0010] An image receiving module, used for receiving noise image data;

[0011] A bias vector and scale selection calculation module, used to calculate bias vector data and scale selection data of the image data using a bias network model;

[0012] A noise reduction calculation module, used for calculating the noise reduction image output using the filter weight and the bias vector data for the noise image data at different scales;

[0013] The fusion module is used to select data according to the scale, fuse the denoised images calculated at different scales into a final denoised image, and output it.

[0014] According to another aspect of the present disclosure, a computing device for image denoising is provided, characterized by comprising:

[0015] Processing device;

[0016] a memory for storing instructions executable by the processing device;

[0017] Wherein, the processing device is configured to implement any one of the above methods when executing instructions.

[0018] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, characterized in that when the computer program instructions are executed by a processing device, any one of the above-mentioned methods is implemented.

[0019] The image denoising method and device disclosed in the present invention adaptively generates scale selection weights and sampling bias information according to the image content, and fuses the multi-scale filtering results to achieve image denoising. Therefore, by generating the optimal sampling information between different scale features in a single frame image, the problem of balancing the image denoising strength and detail loss can be solved, achieving a better denoising effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A block diagram of a computing device for executing an image denoising method according to an embodiment;

[0021] Figure 2 is a flow chart of an image denoising method according to an embodiment;

[0022] Figure 3 A schematic diagram of a U-Net model according to an embodiment;

[0023] Figure 4 is a flow chart of implementing step S230 in one embodiment;

[0024] Figure 5 is a structural block diagram of an image denoising device according to an embodiment;

[0025] Figure 6 is a structural block diagram of a noise reduction calculation module according to an embodiment;

[0026] Figure 7 A schematic diagram of detail region scale selection and bias vector estimation results according to an embodiment;

[0027] Figure 8 A schematic diagram of edge region scale selection and bias vector estimation results according to an embodiment;

[0028] Fig. 9 A schematic diagram of a flat area scale selection and bias vector estimation result according to an embodiment;

[0029] Fig.10 A schematic diagram of a denoised image result after denoising input noisy image data according to an embodiment;

[0030] Fig.11 A block diagram of a computing device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0032] The terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0033] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0034] As used in this specification and claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0035] Figure 1 A block diagram of a computing device 100 for executing an image denoising method according to an embodiment of the present disclosure is shown. Figure 1 The structure of the computing device 100 shown is a simplified illustration. In other embodiments of the present disclosure, the computing device 100 may also adopt other structures for executing the image denoising method.

[0036] The computing device 100 may include a processor 110 , a memory 130 , a network module 150 , and a camera module 170 .

[0037] The processor 110 may be composed of more than one chip, and may include a data analysis and deep learning processor of a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), etc. The processor 110 may read a computer program stored in the memory 130 to execute the image denoising method of an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 may perform calculations for learning of a neural network. The processor 110 may perform calculations for learning of a neural network, such as processing of input data for learning in deep learning (DN), feature extraction in output data, error calculation, and weight value update of a neural network using back propagation. At least one of the CPU, GPGPU, and TPU of the processor 110 may process learning of a network function. For example, the CPU and GPGPU together process learning using a network function and data classification of a network function. Moreover, in an embodiment of the present disclosure, processors of multiple computing devices may be used together to process learning using a network function and data classification of a network function.

[0038] In one embodiment of the present disclosure, the computing device 100 may utilize at least one of a CPU, a GPGPU, and a TPU to distribute and process network functions. In addition, in one embodiment of the present disclosure, the computing device 100 distributes and processes network functions together with other computing devices.

[0039] In one embodiment of the present disclosure, the image processed by the network function may be an image stored in a storage medium of the computing device 100, an image captured by the camera module 170 of the computing device 100, and / or an image transmitted from other computing devices such as an image database through the network module 150. In addition, in one embodiment of the present disclosure, the image processed by the network function may be an image stored in a computer-readable storage medium (for example, including flash memory, etc., but not limited thereto). The computing device 100 may receive an image file stored in a computer-readable storage medium through an input / output interface (not shown).

[0040] The memory 130 may store a computer program for executing the image denoising method according to an embodiment of the present disclosure, and the stored computer program may be read and driven by the processor 110 .

[0041] The network module 150 can be used with other computing devices, servers, etc. to perform data of the image denoising method of an embodiment of the present disclosure. The network module 150 can send and receive image data and other data required by the embodiment of the present disclosure with other computing devices, servers, etc. For example, the network module 150 can receive learning image data in a learning image database, etc. In addition, the network module 150 can enable multiple computing devices to communicate with each other, so as to distribute the learning of the network function among multiple computing devices, and can distribute the data classification using the network function.

[0042] The camera module 170 is a device that can capture an inspection object to generate image data by executing the image denoising method according to an embodiment of the present disclosure. In an embodiment of the present disclosure, the computing device 100 may include more than one camera module 170 .

[0043] Figure 2 A flow chart of an image denoising method 200 according to an embodiment of the present disclosure is shown as follows: Figure 2 As shown, the image denoising method provided by the present disclosure may include the following steps:

[0044] Step S210, receiving noise image data;

[0045] The noise image data in this embodiment may be single-frame image data.

[0046] Step S220, using a bias network model to calculate bias vector data and scale selection data of the image data;

[0047] Specifically, the bias network model can adopt the U-Net model. The U-Net model is an encoder-decoder structure, which has been widely used in the field of deep learning image processing. The convolutional neural network (CNN) implemented by the U-Net model can adaptively generate the optimal scale and its sampling bias information according to the image content to achieve image denoising. Figure 3 , showing the U-Net model used in this embodiment. The input end is single-channel Y component data, and the output end simultaneously outputs k*k*2 channel bias vector data and 4-channel scale selection data, that is, the weights for selecting image sampling data of different scales.

[0048] Step S230, for image data at different scales, using filter weights and the offset vector data to calculate a denoised image output;

[0049] For example, the scale selection data is 9, that is, k is 9, which means that for each pixel in the image, the maximum scale supported is 9*9 pixels. The 9*9 pixels are used as the sampling filter window, and the deformation sampling filter for calculating the denoised image output is shown in formula (1). Where i is the pixel to be processed, is a 9*9 window pixel set centered on i, for The two-dimensional bias vector corresponding to the inner pixel j, w(j) is the filter weight. den_y9(i) is the denoised image output with a scale of 9.

[0050]

[0051] The filter weights may be generated using existing denoising algorithms in the field of deep learning image processing. For example, the ST-PAN (Learning spatial-temporal pixel aggregations for image and video denoising) algorithm may be used to calculate the filter weights.

[0052] Step S240 , according to the scale selection data, the denoised images calculated at different scales are merged into a final denoised image for output.

[0053] For example, this embodiment can also simultaneously generate bias vectors for sampling windows of 3, 5, and 7 scales, and perform corresponding filtering. Finally, the filtering results at scales of 3, 5, 7, and 9 are fused, as shown in formula (2), where w k (i) is the output 4-channel scale selection data, that is, the selection weights of image sampling data of different scales, which has the same spatial resolution as the input noisy image. den_y(i) is the final output denoised image data.

[0054]

[0055] According to the image denoising method 200 in the embodiment of the present disclosure, for detail-free areas in noisy images, a large-scale deformation sampling unit is used to fuse more pixels for effective noise reduction processing; and for detail areas, a small-scale deformation sampling unit can reduce the risk of bias vector prediction and reduce the noise reduction loss in detail areas. Scale selection weights and sampling bias information are adaptively generated according to the image content, and the multi-scale filtering results are fused to achieve image denoising. Therefore, by generating the optimal sampling information between different scale features in a single frame image, the problem of balancing image denoising intensity and detail loss can be solved to achieve a better noise reduction effect.

[0056] Figure 4A flowchart of implementing step S230 according to an embodiment of the present disclosure is shown. Figure 4 As shown, the implementation step S230 provided by the present disclosure may include the following steps:

[0057] S2301, for the image data at a small scale, using the bias vector data of the large scale image data;

[0058] S2302: For the image data at a small scale, a denoised image output is calculated using a filter weight and the bias vector data of the large scale image data.

[0059] According to this embodiment, for the small-scale image data deformation sampling unit, it is not necessary to generate the corresponding offset vector data additionally, but the offset vector data required by the small-scale sampling unit is obtained by sharing the existing maximum-scale offset data. Taking 5*5 sampling filtering as an example, the corresponding filtering is shown in formula (3):

[0060]

[0061] Figure 5 A structural block diagram of an image denoising device 500 according to an embodiment of the present disclosure is shown as follows: Figure 5 As shown, the image denoising device 500 provided by the present disclosure may include the following modules:

[0062] An image receiving module 501 is used to receive noise image data;

[0063] The noise image data in this embodiment may be single-frame image data.

[0064] A bias vector and scale selection calculation module 502, used to calculate bias vector data and scale selection data of the image data using a bias network model;

[0065] Specifically, the bias network model can adopt the U-Net model. The U-Net model is an encoder-decoder structure, which has been widely used in the field of deep learning image processing. The convolutional neural network (CNN) implemented by the U-Net model can adaptively generate the optimal scale and its sampling bias information according to the image content to achieve image denoising. Figure 3 , showing the U-Net model used in this embodiment. The input end is single-channel Y component data, and the output end simultaneously outputs k*k*2 channel bias vector data and 4-channel scale selection data.

[0066] A noise reduction calculation module 503 is used to calculate the noise reduction image output using the filter weights and the offset vector data for the image data at different scales;

[0067] For example, the scale selection data is 9, that is, k is 9, which means that for each pixel in the image, the maximum scale supported is 9*9 pixels. The 9*9 pixels are used as the sampling filter window, and the deformation sampling filter for calculating the denoised image output is shown in formula (1). Where i is the pixel to be processed, is a 9*9 window pixel set centered on i, for The two-dimensional bias vector corresponding to the inner pixel j, w(j) is the filter weight. den_y9(i) is the denoised image output with a scale of 9.

[0068]

[0069] The filter weights may be generated using existing denoising algorithms in the field of deep learning image processing. For example, the ST-PAN (Learning spatial-temporal pixel aggregations for image and video denoising) algorithm may be used to calculate the filter weights.

[0070] A fusion module 504 is used to select data according to the scale, fuse the denoised images calculated at different scales into a final denoised image, and output it;

[0071] For example, this embodiment can also simultaneously generate bias vectors for sampling windows of 3, 5, and 7 scales, and perform corresponding filtering. Finally, the filtering results at scales of 3, 5, 7, and 9 are fused, as shown in formula (2), where w k (i) is the weight for the output 4-channel scale selection, which has the same spatial resolution as the noisy image. den_y(i) is the final output denoised image data.

[0072]

[0073] According to the image denoising device 500 in the embodiment of the present disclosure, for detail-free areas in the noisy image, a large-scale deformation sampling unit is used to fuse more pixels for effective noise reduction processing; and for detail areas, a small-scale deformation sampling unit can reduce the risk of bias vector prediction and reduce the noise reduction loss in detail areas. Scale selection weights and sampling bias information are adaptively generated according to the image content, and the multi-scale filtering results are fused to achieve image denoising. Therefore, by generating the optimal sampling information between different scale features in a single frame image, the problem of balancing image denoising intensity and detail loss can be solved to achieve a better noise reduction effect.

[0074] Figure 6A structural block diagram of the noise reduction calculation module 503 according to an embodiment of the present disclosure is shown as follows: Figure 6 As shown, the noise reduction calculation module 503 provided by the present disclosure may include the following steps:

[0075] The bias vector selection module 5031 uses the bias vector data of the large-scale image data for the image data at the small scale;

[0076] The denoised image calculation module 5032 calculates the denoised image output using the filter weights and the bias vector data of the large-scale image data for the image data at the small scale.

[0077] According to this embodiment, for the small-scale image data deformation sampling unit, it is not necessary to generate the corresponding offset vector data additionally, but the offset vector data required by the small-scale sampling unit is obtained by sharing the existing maximum-scale offset data. Taking 5*5 sampling filtering as an example, the corresponding filtering is shown in formula (3):

[0078]

[0079] Figures 7 to 9 The image results of the scale selection data generated according to the characteristics of different regions and the deformation sampling bias vector data within each scale using the image denoising method disclosed in the present invention are shown.

[0080] in Figure 7 Schematic diagram of detail area scale selection and bias vector estimation results; Figure 8 Schematic diagram of edge region scale selection and bias vector estimation results; Fig. 9 Schematic diagram of flat area scale selection and bias vector estimation results.

[0081] Fig.10 The denoised image is a schematic diagram of the denoised image result after denoising the input noisy image data using the image denoising method disclosed in the present invention. (a) is a noisy image, and (b) is a denoised image after denoising the input noisy image data using the image denoising method disclosed in the present invention.

[0082] Fig.11 A block diagram of a computing device according to an embodiment of the present disclosure. Fig.11 A simple and general schematic diagram of an exemplary computing device that can embody the embodiments of the present disclosure is shown. The present disclosure does not limit the implementation methods, and various methods such as digital circuits and analog circuits can be used to implement, for example, application-specific integrated circuits ASIC, FPGA, etc. can be used to implement.

[0083] As described above, generally, the present disclosure is related to computer-executable instructions that can be executed on more than one computer, and anyone skilled in the art to which the present disclosure belongs can understand that the present disclosure can be combined with other program modules and / or a combination of hardware and software.

[0084] Typically, a program module includes routines, programs, components, data structures, and others that execute specific characters or embody specific abstract data types. In addition, as long as one is a person of ordinary skill in the art to which the present disclosure belongs, it can be known that the method of the present disclosure can be implemented by other computer system structures including single processor or multi-processor computer systems, microcomputers, central computers, personal computers, handheld computing devices, microprocessor-based or programmable household appliances, and others (which can be connected to more than one related device to work).

[0085] The embodiments described in the present disclosure may also be implemented in a distributed computing environment where certain tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both logical and remote memory storage devices.

[0086] Typically, a computer includes a variety of computer-readable media. Any medium that can be accessed by a computer can be a computer-readable medium, which includes volatile media and non-volatile media, temporary (transitory) media and non-transitory (non-transitory) media, mobile media and non-mobile media. As a non-limiting embodiment, a computer-readable medium may include a computer-readable storage medium and a computer-readable transmission medium. Computer-readable storage media include volatile media and non-volatile media, temporary media and non-transitory media, mobile media and non-mobile media embodied by any method or technology of storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other storage technology, CD-read-only memory, digital optical disk (DVD, digital video disk) or other optical disk storage devices, magnetic tapes, magnetic tapes, magnetic disk storage devices or other magnetic storage devices or any other medium that can be accessed by a computer and stores the required data, but is not limited to this.

[0087] Generally, computer-readable recording media embody computer-readable instructions, data structures, program modules or other data in a non-modulated data signal such as a carrier wave or other transmission mechanism and include all information transmission media. Non-modulated data signals are signals that set or change one or more characteristics of the signal in a manner that encodes the information in the signal. As non-limiting examples, computer-readable transmission media include wired media such as wired networks or direct-wired connections, audio, RF, infrared and other wireless media. Any combination of the above media is also included in the scope of computer-readable transmission media.

[0088] Fig.11 The exemplary environment 1100 includes a computer 1102, which includes a processing device 1104, a system memory 1106, and a system bus 1108. The system bus 1108 connects system components including the system memory 1106 (but not limited to this) to the processing device 1104. The processing device 1104 can be any of a variety of commonly used processors. Dual processors and other multi-processors can also be used as the processing device 1104.

[0089] The system bus 1108 may be one of a plurality of types of bus structures using a logical bus that can be connected to a storage bus, a peripheral device bus, and any of a plurality of commonly used buses in addition. The system memory 1106 includes a non-volatile memory 1110 and a random access memory 1112. The basic input / output system (BIOS) is stored in a non-volatile memory 1110 such as a read-only memory, an electronic program control read-only memory, or an electrically erasable programmable read-only memory, and the input / output system includes a basic program for transmitting information between structural elements in the computer 1102 when starting. The random access memory 1112 may also include a high-speed random access memory such as an appropriate random access memory for caching data.

[0090] The computer 1102 also includes an internal hard disk drive 1114 (HDD) (e.g., EIDE, SATA), a floppy disk drive 1116 (FDD) (e.g., reading from or recording on a removable floppy disk 1118), and an optical disk drive 1120 (e.g., reading from a CD-ROM disk 1122 or reading from or recording on other high-capacity optical media such as a DVD). The internal hard disk drive 1114 can also be used for external purposes in an appropriate frame (not shown). The hard disk drive 1114, the magnetic disk drive 1116, and the optical disk drive 1120 can be connected to the system bus 1108 through a hard disk drive interface 1124, a magnetic disk drive interface 1126, and an optical drive interface 1128, respectively. The interface 1124 for embodying an external drive includes at least one or both of a Universal Serial Bus (USB) and an IEEE 1394 interface technology.

[0091] These drives and computer-readable media connected thereto provide non-volatile storage of data, data structures, computer-executable instructions, and the like. In the case of computer 1102, the drives and media store any data in an appropriate digital form. In the description of the above-mentioned computer-readable media, hard disk drives, removable floppy disks, and removable optical media such as CDs or DVDs, as well as other types of media that can be read by computers, such as magnetic tapes, flash memory cards, disk cartridges, and the like, which are well known to those of ordinary skill in the art to which the present disclosure belongs, can also be used in the exemplary operating environment, and any media can include computer-executable instructions for performing the methods of the present disclosure.

[0092] A plurality of program modules including an operating system 1130, one or more application programs 1132, other program modules 1134, and program data 1136 may be stored in the drives and random access memory 1112. All or a portion of the operating system, applications, modules, and / or data may also be cached in random access memory 1112. The present disclosure may be embodied in a plurality of commercially available operating systems or combinations of operating systems.

[0093] The user can input commands and information to the computer 1102 through one or more wired or wireless input devices, such as a keyboard 1138 and a mouse 1140. Other input devices (not shown) can be microphones, IR remote controls, joysticks, game pads, styluses, touch screens, and others. These and other input devices are generally connected to the processing device 1104 through an input device interface 1142 connected to the system bus 1108, and can also be connected through a parallel port, an IEEE 1394 serial port, a game port, a USB port, an IR interface, and other interfaces.

[0094] A display 1144 or other type of display device is also connected to the system bus 1108 via an interface such as a video adapter 1146. In addition to the display 1144, computers typically include speakers, printers, and other peripheral output devices (not shown).

[0095] The computer 1102 can operate in a networked environment using logical connections to one or more remote computers, such as multiple remote computers 1148, through wired and / or wireless communications. Multiple remote computers 1148 can be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and typically include multiple or all of the structural elements described for the computer 1102, and for simplicity, only a memory storage device 1150 is shown. The logical connections shown include a near field communication network 1152 (LAN) and / or a larger network, such as a wired or wireless connection to a long-distance communication network 1154 (WAN). Such near field communication network and long-distance communication network network environments are commonly used in firms and companies to reduce the burden on enterprise-wide computer networks such as intranets, which can be connected to a worldwide computer network, such as the Internet.

[0096] When used in a near field communication network environment, the computer 1102 is connected to the local network 1152 through a wired and / or wireless communication network interface or adapter 1156. The adapter 1156 facilitates wired or wireless communication with the near field communication network 1152, which may also include a wireless access point provided thereto for communication with the wireless adapter 1156. When used in a remote communication network environment, the computer 1102 may include a modem 1158, or other means for connecting to a communication computing device on the remote communication network 1154, or for establishing communication with the remote communication network 1154 via the Internet or the like. The modem 1158, which may be internal or external and wired or wireless, is connected to the system bus 1108 via the serial port interface 1142. In a networked environment, the plurality of program modules described for the computer 1102, or portions thereof, may be stored in a remote memory storage device 1150. The network connection shown is illustrative, and other means for establishing communication links between multiple computers may be used.

[0097] The computer 1102 communicates with any wireless device or individual that operates via a wireless communication configuration, such as a printer, scanner, desktop and / or portable computer, a personal data assistant (PDA), a communication satellite, any device or location associated with a wirelessly detectable tag, and a telephone. This includes at least Wi-Fi and Bluetooth wireless technologies. Thus, as described above, the communication is a predefined structure or simply an ad hoc communication between at least two devices.

[0098] Wireless Fidelity (Wi-Fi) allows connections to the Internet, etc., without wires. Wi-Fi is a wireless technology that enables devices such as computers to send and receive data, such as cell phones, both indoors and outdoors, i.e., anywhere within the reach of a base station. Wi-Fi networks are secure and reliable, and use a wireless technology called IEEE 802.11 (a, b, g, others) to provide high-speed wireless connections. Wi-Fi can be used to connect computers to the Internet and wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks operate in the unlicensed 2.4 and 5 GHz wireless frequencies, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or can operate in products that include two frequencies (dual-band).

[0099] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1110 including computer program instructions that can be executed by a processing component 1104 of a computer 1102 to perform the above method.

[0100] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should know that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0101] It should be further explained that although Figure 2 , Figure 4 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 , Figure 4 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0102] It should be understood that the above-mentioned device embodiments are only illustrative, and the device disclosed herein may also be implemented in other ways. For example, the division of units / modules described in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components may be combined, or may be integrated into another system, or some features may be ignored or not executed.

[0103] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present disclosure may be integrated into one unit / module, or each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.

[0104] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. Unless otherwise specified, the artificial intelligence processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, an ASIC, etc. Unless otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc.

[0105] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution disclosed herein is essentially or partly contributed to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a memory, including several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the disclosure. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0106] In the above embodiments, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The foregoing content can be better understood in accordance with the following terms:

[0108] Clause A1. A data denoising method, characterized in that the method comprises:

[0109] receiving noisy image data;

[0110] Calculating bias vector data and scale selection data of the image data using a bias network model;

[0111] For noisy image data at different scales, the filter weights and the bias vector data are used to calculate the denoised image output;

[0112] According to the scale selection data, the denoised images calculated at different scales are fused into a final denoised image for output.

[0113] Clause A2. A method according to clause A1, characterized in that the selecting data according to the scale, for image data at different scales, using filtering weights and the bias vector data to calculate the denoised image output, comprises:

[0114] For image data at a small scale, the bias vector data of the large scale image data is used;

[0115] For image data at a small scale, the denoised image output is calculated using filter weights and the bias vector data of the large scale image data.

[0116] Clause A3. The method according to clause A1, characterized in that:

[0117] The bias network model is a U-Net model.

[0118] Clause A4. The method according to clause A1, characterized in that:

[0119] The different scales include scales 3, 5, 7, and 9.

[0120] Item A5. A data denoising device, characterized in that the device comprises:

[0121] An image receiving module, used for receiving noise image data;

[0122] A bias vector and scale selection calculation module, used to calculate bias vector data and scale selection data of the image data using a bias network model;

[0123] A noise reduction calculation module, used for calculating the noise reduction image output using the filter weight and the bias vector data for the noise image data at different scales;

[0124] The fusion module is used to select data according to the scale, fuse the denoised images calculated at different scales into a final denoised image, and output it.

[0125] Clause A6. The device according to clause A5, wherein the noise reduction calculation module further comprises:

[0126] A bias vector selection module uses bias vector data of large-scale image data for image data at a small scale;

[0127] The denoised image calculation module calculates the denoised image output using the filter weights and the bias vector data of the large-scale image data for the image data at a small scale.

[0128] Clause A7. The device according to clause A5, characterized in that:

[0129] The bias network model is a U-Net model.

[0130] Clause A8. The device according to clause A5, characterized in that:

[0131] The different scales include scales 3, 5, 7, and 9.

[0132] Clause A9. A computing device for image denoising, comprising:

[0133] Processing device;

[0134] a memory for storing instructions executable by the processing device;

[0135] Wherein, the processing device is configured to implement the method described in any one of claims 1 to 4 when executing instructions.

[0136] Clause A10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that when the computer program instructions are executed by a processing device, the method described in any one of claims 1 to 4 is implemented.

[0137] The embodiments of the present disclosure are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present disclosure. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present disclosure. At the same time, changes or deformations made by those skilled in the art based on the ideas of the present disclosure, the specific implementation methods and the scope of application of the present disclosure, all belong to the scope of protection of the present disclosure. In summary, the content of this specification should not be understood as a limitation on the present disclosure.

Claims

1. An image denoising method, applied to a computing device, characterized in that: The method comprises: receiving noisy image data; Using a bias network model to calculate bias vector data and scale selection data of the image data, and obtain selection weights of image sampling data of different scales, wherein the scale selection data can represent a maximum scale of the number of pixels supported by the sampling filter for each pixel in the image; For the noise image data at different scales, a sampling filtering window is selected according to the scale selection data, and a noise reduction image output is calculated using a filtering weight and the offset vector data, wherein the filtering weight is used to assign a weight to each pixel value in the window, the filtering weight is generated by a denoising algorithm, and the offset vector data is a two-dimensional offset vector corresponding to a pixel point in the window, and is used to offset the position of each pixel point in the window; According to the selection weights of the image sampling data at different scales, the denoised images calculated at different scales are fused into a final denoised image for output.

2. The method according to claim 1, characterized in that For noisy image data at different scales, a sampling filtering window is selected according to the scale selection data, and a noise reduction image output is calculated using a filtering weight and the offset vector data, including: For image data at a small scale, the bias vector data of the large scale image data is used; For image data at a small scale, the denoised image output is calculated using filter weights and the bias vector data of the large scale image data.

3. The method according to claim 1, characterized in that: The bias network model is a U-Net model.

4. The method according to claim 1, characterized in that: The different scales include scales 3, 5, 7, and 9.

5. An image denoising device, characterized in that: The device comprises: An image receiving module, used for receiving noise image data; A bias vector and scale selection calculation module, used to calculate bias vector data and scale selection data of the image data using a bias network model to obtain selection weights of image sampling data of different scales, wherein the scale selection data can represent a maximum scale of the number of pixels supported by the sampling filter for each pixel in the image; A noise reduction calculation module, for selecting a sampling filtering window for noise image data at different scales according to the scale selection data, and calculating a noise reduction image output using a filtering weight and the offset vector data, wherein the filtering weight is used to assign a weight to each pixel value in the window, the filtering weight is generated by a denoising algorithm, and the offset vector data is a two-dimensional offset vector corresponding to a pixel point in the window, and is used to offset the position of each pixel point in the window; The fusion module is used to fuse the denoised images calculated at different scales into a final denoised image according to the selection weights of the image sampling data at different scales, and output it.

6. The device according to claim 5, characterized in that The noise reduction calculation module also includes: A bias vector selection module uses bias vector data of large-scale image data for image data at a small scale; The denoised image calculation module calculates the denoised image output using the filter weights and the bias vector data of the large-scale image data for the image data at a small scale.

7. The device according to claim 5, characterized in that Features: The bias network model is a U-Net model.

8. The device according to claim 5, characterized in that: The different scales include scales 3, 5, 7, and 9.

9. A computing device for image denoising, characterized in that: include: Processing device; a memory for storing instructions executable by the processing device; Wherein, the processing device is configured to implement the method described in any one of claims 1 to 4 when executing instructions.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processing device, the method according to any one of claims 1 to 4 is implemented.

Images