CN118052727A - Image noise reduction processing method and device and computer equipment - Google Patents

Abstract

The present application relates to the field of image processing, and in particular, to an image noise reduction processing method, an image noise reduction processing device, and a computer device. The image noise reduction processing method comprises the following steps: scaling the original image to be denoised according to the scaling coefficient sequence, quantizing the brightness value of the target pixel point in the scaled image, inversely scaling the quantized image according to the scaling coefficient sequence to obtain a restored image, inputting a plurality of restored images into the image denoising model, and performing weighted modulation on the result image output by the model according to the denoising effect to obtain the final denoising image. Therefore, the frequency adjustment of the image information can be realized through scaling and quantization, the restored images of the original image in different frequency bands are obtained, the noise reduction results of the multiple restored images are modulated into the noise reduction image, the noise reduction of the information in different frequency bands can be realized under the condition that the noise reduction model is not changed, the noise reduction effect of the dynamic frequency division adjusting image is realized, and the adjustability and the applicability of the noise reduction model are improved.

Description

Image noise reduction processing method, device and computer equipment

Technical Field

The present application relates to the field of image processing, and in particular to an image noise reduction processing method, device and computer equipment.

Background technique

Image denoising is an image processing method that can reduce image noise and improve image quality. It can mainly include denoising based on combined filters and denoising based on neural networks. The filter-based denoising method is more hardware-friendly, but the denoising frequency is single, and stacking multiple filters can easily cause loss of details; the neural network-based denoising method is less hardware-friendly, but through complex network learning, it can better handle image noise while ensuring details.

In order to ensure that the details of the image are not seriously damaged, the filter-based denoising model often uses a single-band filter, which cannot effectively remove noise in non-bands. However, due to its high parameter and computational requirements, the neural network denoising model often needs to reduce the number of parameters and computing power in order to obtain a higher hardware matching degree. As a result, the neural network denoising model cannot process numerical information in the full frequency domain, and can only use computing power to process values concentrated in the frequency domain. These two methods have strong image denoising capabilities in a specific frequency band, but have poor processing capabilities for image data outside of this specific frequency band, resulting in the overall low usability of the image denoising model.

Summary of the invention

The embodiments of the present application provide an image denoising processing method, apparatus and computer equipment to solve the problem that the image denoising model has a single denoising frequency band and poor denoising capability in all frequency bands.

In a first aspect, an embodiment of the present application provides an image noise reduction processing method, comprising:

Scaling the original image according to a scaling coefficient sequence to obtain a plurality of scaled images; the scaling coefficient sequence includes a plurality of scaling constants;

quantizing the brightness values of target pixels in the scaled image to obtain a plurality of quantized images corresponding to the scaled image;

Inversely scale the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images;

Inputting a plurality of restored images into a preset image denoising model, and obtaining a plurality of result images output by the image denoising model; the image denoising model is a denoising model constructed based on a filter or a denoising model constructed based on a neural network;

Determine a modulation coefficient of the result image according to the noise reduction effect of the result image;

The multiple result images are weighted and combined according to the modulation coefficient to obtain a denoised image.

In a possible implementation, scaling the original image according to the scaling coefficient sequence to obtain a plurality of scaled images includes:

Obtaining an image to be processed and a preset zoom factor sequence;

Map the brightness value of each pixel in the image to be processed to the target interval to obtain the original image;

The brightness values of the pixels in the original image are multiplied by the scaling constants in the scaling coefficient sequence to obtain multiple scaled images.

In one possible implementation, the target interval is [0,1].

In a possible implementation, quantizing the brightness value of the target pixel in the scaled image includes performing one or more combinations of rounding down, rounding up, and rounding off on the brightness value of the target pixel.

In a possible implementation, determining the modulation coefficient of each result image according to the noise reduction effect of the result image includes:

Determine the denoising frequency band of the image denoising model according to the denoising effect of the result image;

The noise frequency band of the result image is calculated according to the brightness value of the pixel points in the result image, and the overlapping frequency bands of the noise frequency band and the noise reduction frequency band are counted;

The resulting image is assigned a modulation coefficient based on the proportion of the overlapping frequency band to the noise frequency band.

In a possible implementation, determining the modulation coefficient of the result image according to the noise reduction effect of the result image includes:

Calculate the signal-to-noise ratio of the resulting image;

The resulting image is assigned a modulation coefficient based on the signal-to-noise ratio.

In a possible implementation, determining the modulation coefficient of each result image according to the noise reduction effect of the result image includes:

Get high-definition images of the original images;

The high-definition image and the multiple result images are sequentially input into a pre-trained weight allocation model; the weight allocation model is a neural network model for allocating corresponding weight coefficients according to the noise reduction effect of the input image;

Get the modulation coefficient corresponding to the result image output by the weight distribution model.

In a possible implementation manner, after the plurality of result images are weighted and combined according to the modulation coefficients to obtain the noise-reduced image, the method further includes:

Input the original image into the image denoising model;

Obtaining an original denoised image corresponding to the original image output by the image denoising model;

The target denoised image is obtained by combining the original denoised image and the denoised image in a linear manner.

In a second aspect, the present application provides an image noise reduction processing device, comprising:

A scaling module, used for scaling the original image according to a scaling coefficient sequence to obtain a plurality of scaled images; the scaling coefficient sequence includes a plurality of constants;

A quantization module, used to quantize the brightness value of the target pixel in the scaled image to obtain a plurality of quantized images corresponding to the scaled image;

A restoration module, used for inversely scaling the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images;

An input module, used to input a plurality of restored images into a preset image denoising model, and obtain a plurality of result images output by the image denoising model; the denoising model is a denoising model constructed based on a filter or a denoising model constructed based on a neural network;

An acquisition module, used for determining a modulation coefficient of each result image according to a noise reduction effect of the result image;

The image modulation module is used to weight and combine multiple result images according to the modulation coefficient to obtain a noise-reduced image.

In a third aspect, the present application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor calls and executes the computer program from the memory, the steps of the image denoising method described in the first aspect are implemented.

In a fourth aspect, the present application provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the image denoising method described in the first aspect above.

In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the image denoising method described in the first aspect.

Through the technical solution provided by the present application, the original image to be denoised can be scaled according to the scaling coefficient sequence, and the brightness value of the target pixel in the scaled image can be quantized, and then the quantized image can be inversely scaled according to the scaling coefficient sequence to obtain a restored image. After multiple restored images are input into the image denoising model, the result image output by the model can be weighted modulated according to the denoising effect to obtain the final denoised image. In this way, the frequency of the image information can be adjusted by scaling and quantizing, and the restored images of the original image in different frequency bands can be obtained. Then, the denoising results of the multiple restored images can be modulated into denoised images. The information of different frequency bands can be denoised without changing the denoising model, and the image denoising effect can be adjusted by dynamic frequency division, thereby improving the adjustability and applicability of the denoising model.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

FIG1 is a schematic diagram of a process flow of an image noise reduction method provided in an embodiment of the present application;

FIG2 is a schematic diagram of a process of scaling an original image according to a scaling coefficient sequence in an embodiment of the present application;

FIG3 is a schematic diagram of a flow chart of determining a modulation coefficient according to a noise reduction effect of a result image in an embodiment of the present application;

FIG4 is a schematic diagram of a flow chart of determining a modulation coefficient according to a noise reduction effect of a result image provided in another embodiment of the present application;

FIG5 is a schematic diagram of a flow chart of determining a modulation coefficient according to a noise reduction effect of a result image provided in another embodiment of the present application;

FIG6 is a schematic diagram of a flow chart of an image noise reduction method provided in another embodiment of the present application;

FIG7 is a schematic diagram showing a comparison of noise reduction effects of an image noise reduction processing method provided in an embodiment of the present application;

FIG8 is a structural block diagram of an image noise reduction processing device provided in an embodiment of the present application;

FIG. 9 is a structural block diagram of a computer device provided in an embodiment of the present application.

Detailed ways

To facilitate the technical solution of the application, some concepts involved in the application are first explained below.

It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their ordinary and common meanings.

The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

At present, there are two main approaches to image denoising: one is to use a filter-based denoising model, and the other is to use a neural network-based denoising model. Both methods have their own advantages. The filter-based denoising model is easier to combine with hardware and is simpler to apply. The neural network-based denoising model can obtain stronger denoising capabilities for a certain type of image through machine learning.

However, both filter-based and neural network-based noise reduction models need to be combined with hardware devices in their deployment and application. Due to the limitations of most hardware devices, the number of filters that can be stacked is limited, and the computing power resources that can be provided are limited. As a result, both noise reduction models can only bias the noise reduction capability to a fixed frequency band, and the noise reduction effect is poor outside this fixed frequency band.

In response to the above technical problems, the embodiments of the present application provide an image denoising processing method, apparatus and computer equipment, which can improve the full-band denoising processing capability of the image denoising model and enhance the adjustability of the image denoising model.

FIG1 is a flow chart of an image noise reduction processing method provided by an embodiment of the present application, wherein the processing method comprises:

Step S210: scaling the original image according to the scaling coefficient sequence to obtain a plurality of scaled images.

Wherein, the scaling coefficient sequence includes a plurality of scaling constants. The scaling constant can be any positive real number, and the original image can be an image that needs to be subjected to noise reduction processing. In some embodiments, the scaling constant can be obtained by reading a storage medium or by communicating with a host computer, or can be randomly generated. In some other embodiments, the original image can be obtained by reading a storage medium or by communicating with a host computer, or can be obtained by an image acquisition device such as a camera or a video camera.

In some possible implementations, the scaling constant in the scaling coefficient sequence may be multiplied by the brightness value of the pixel in the original image to obtain the scaled image. For example, if the scaling coefficient sequence includes three scaling constants L, M, and N, the brightness value of each pixel in the original image is multiplied by L to obtain a scaled image corresponding to L, and then the same operation is repeated using M and N, and finally three scaled images corresponding to L, M, and N are obtained.

Step S220 , quantizing the brightness values of target pixels in the scaled image to obtain a plurality of quantized images corresponding to the scaled image.

The target pixel points may be all the pixel points of the zoomed image, or may be the pixel points selected according to a preset brightness threshold.

In some possible implementations, quantization is performed on the brightness value of the pixel point, which may be performed by rounding down the brightness value, rounding up the brightness value, or rounding the brightness value.

In some other possible implementations, different quantization operations can be performed on the brightness values of pixel points by setting quantization thresholds, for example, rounding down brightness values that are higher than or equal to the first quantization threshold, rounding up brightness values that are higher than or equal to the second quantization threshold and lower than the first quantization threshold, and rounding up brightness values that are higher than the third quantization threshold and lower than the second quantization threshold.

In some other possible implementations, a brightness threshold may be preset, and pixels whose brightness values are greater than or equal to the brightness threshold are target pixels that need to be quantized.

Step S230 , inversely scale the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images.

In some possible implementations, multiple scaling constants in the scaling coefficient sequence may be used to sequentially perform inverse operations on the quantized image. For example, the reciprocal of the scaling constant may be multiplied by the brightness value of a pixel in the quantized image to obtain a corresponding restored image.

Since the brightness values of the pixels in the original image are quantized after scaling, the brightness values of some pixels change after quantization. This makes the original image and the restored image obtained after inverse scaling have the same size, but the brightness values of the pixels in the original image are different from the brightness values of the pixels in the restored image, so that the restored image has a different frequency band from the original image. In this way, through a series of operations of scaling, quantization, and inverse scaling, the restored images corresponding to the original image in different frequency bands can be obtained.

Step S240: input the multiple restored images into a preset image denoising model, and obtain multiple result images output by the image denoising model.

The image denoising model may be a denoising model constructed based on a filter or a denoising model constructed based on a neural network.

In some possible implementations, a plurality of restored images may be sequentially input into an image denoising model, and the image denoising model may perform denoising processing on different restored images respectively to obtain a plurality of denoised result images.

Step S250: determining a modulation coefficient of the result image according to the noise reduction effect of the result image.

Since the image denoising model often only has the denoising capability for a fixed frequency band, and even within the fixed frequency band, the denoising capabilities for different frequency intervals may be different, the denoising effects of the resulting images obtained by the image denoising model after denoising different restored images are also different.

In some possible implementations, a modulation coefficient may be determined for each result image by comparing the noise reduction effects of different result images, and the modulation coefficient is used to represent the weight of the result image in the subsequent modulation combination process. It is understandable that, in order to obtain an image with less noise, a higher modulation coefficient may be assigned to a result image with better noise reduction effect, and a lower modulation coefficient may be assigned to a result image with poor noise reduction effect.

Step S260: weightedly combine multiple result images according to the modulation coefficients to obtain a denoised image.

In some possible implementations, each result image may be multiplied by a corresponding modulation coefficient and then superimposed to obtain a noise-reduced image that integrates multiple result images.

In the above embodiment, the original image to be denoised can be scaled according to the scaling coefficient sequence, and the brightness value of the target pixel in the scaled image can be quantized, and then the quantized image can be inversely scaled according to the scaling coefficient sequence to obtain a restored image, and after multiple restored images are input into the image denoising model, the result image output by the model can be weighted modulated according to the denoising effect to obtain the final denoised image. In this way, the frequency adjustment of the image information can be achieved through scaling and quantization, and the restored images of the original image in different frequency bands can be obtained, and then the denoising results of the multiple restored images can be modulated into denoised images, and the information of different frequency bands can be denoised without changing the denoising model, so as to achieve the dynamic frequency division adjustment image denoising effect, and improve the adjustability and applicability of the denoising model.

FIG. 2 is a schematic diagram of a process of scaling an original image according to a scaling coefficient sequence in an embodiment of the present application. As shown in FIG. 2 , the process includes:

Step S2102, obtaining an image to be processed and a preset zoom coefficient sequence.

The image to be processed may be an image that needs to be subjected to noise reduction processing.

In some possible implementations, the scaling coefficient sequence includes multiple scaling constants, and the scaling constants can be obtained in the following manner: using several randomly generated constants to perform the above steps S210, S220, S230, and S240 on a pre-prepared noisy image, and comparing the denoising results corresponding to different constants; if the noise area in the denoising result corresponding to a certain constant changes significantly relative to the noisy image, then the constant can be used as the scaling constant of the scaling coefficient sequence.

Step S2104, mapping the brightness value of each pixel in the image to be processed to the target interval to obtain the original image.

In some possible implementations, the target interval may be [0, 1].

In some possible implementations, the image to be processed may be normalized, and the value range of the brightness value of each pixel in the image to be processed may be mapped to [0, 1] to obtain the original image after value range compression.

Step S2106: multiply the brightness values of the pixels in the original image by the scaling constants in the scaling coefficient sequence to obtain a plurality of scaled images.

In some possible implementations, the scaling constant may be a constant greater than 1.

In the above embodiment, by mapping the value range of the brightness value of the pixel points in the image to be processed to the target interval, the brightness value range of different images can be unified, the information loss of the image during scaling and quantization can be reduced, and the image noise reduction processing effect can be enhanced.

FIG3 is a schematic diagram of a process for determining a modulation coefficient according to a noise reduction effect of a result image in an embodiment of the present application. As shown in FIG3 , the process includes:

Step S2501, determining the denoising frequency band of the image denoising model according to the denoising effect of the result image.

In some possible implementations, the brightness change values of the same pixel points in the result image and the restored image can be counted. For example, the brightness values of the characteristic pixel points in the result image and the corresponding pixel points in the restored image can be subtracted, and the highest value is taken from the differences between the different result images and the restored image at the same pixel point. The frequency of the pixel point in the restored image corresponding to the highest value is used as the denoising frequency of the image denoising model. By setting multiple characteristic pixel points, the upper and lower limits of the denoising frequency of the image denoising model can be determined, thereby obtaining the denoising frequency band. Since the image denoising model performs denoising on image information in a fixed frequency band, the denoising frequency band of the image denoising model can be determined according to the brightness change of the pixel points in the result image and the pixel points in the restored image.

Step S2502: Calculate the noise frequency band of the result image according to the brightness values of the pixels in the result image, and count the overlapping frequency bands of the noise frequency band and the noise reduction frequency band.

Step S2503, allocating a modulation coefficient to the result image according to the ratio of the overlapping frequency band to the noise frequency band.

In some possible implementations, the higher the ratio of the overlapping frequency band to the noise frequency band, the larger the modulation coefficient allocated to the result image, and the lower the ratio of the overlapping frequency band to the noise frequency band, the smaller the modulation coefficient allocated to the result image.

In the above embodiment, the noise reduction frequency band of the image noise reduction model can be determined by the brightness change between the result image and the restored image, and then according to the overlapping ratio of the noise reduction frequency band and the noise frequency band in the result image, a larger modulation coefficient can be allocated to the result image with a higher ratio, thereby improving the noise reduction effect of the denoised image after modulation.

FIG4 is a schematic diagram of a process for determining a modulation coefficient according to a noise reduction effect of a result image in another embodiment of the present application. As shown in FIG4 , the process includes:

Step S2504, calculating the signal-to-noise ratio of the result image.

In some possible implementations, the local variance of all pixels in the result image may be calculated, the maximum value of the local variance may be taken as the signal variance, the minimum value of the local variance may be taken as the noise variance, and the ratio of the signal variance to the noise variance may be taken as the signal-to-noise ratio of the result image.

Step S2505: assign a modulation coefficient to the result image according to the signal-to-noise ratio.

In some possible implementations, the greater the signal-to-noise ratio of the result image, the greater the modulation coefficient allocated to the result image, and the smaller the signal-to-noise ratio, the smaller the modulation coefficient allocated to the result image.

In the above embodiment, the image signal-to-noise ratio can be quickly calculated through the local variance of the pixels in the result image, and the modulation coefficient of the result image is allocated according to the signal-to-noise ratio, which can speed up the modulation process and improve the image noise reduction processing efficiency.

FIG5 is a schematic diagram of a process for determining a modulation coefficient according to a noise reduction effect of a result image in another embodiment of the present application. As shown in FIG5 , the process includes:

Step S2506, obtaining a high-definition image of the original image.

In some possible implementations, Gaussian filtering may be performed on the original image to obtain a filtered high-definition image.

Step S2507, inputting the high-definition image and multiple result images into the pre-trained weight allocation model in sequence.

Among them, the weight allocation model is a neural network model used to allocate corresponding weight coefficients according to the noise reduction effect of the input image.

In some possible implementations, a noisy image and multiple denoised images may be used as inputs in advance to train a neural network model, so that the neural network model is assigned different weight coefficients according to the image denoising effect.

Step S2508, obtaining the modulation coefficient corresponding to the result image output by the weight distribution model.

In some other implementations, when executing the above step S2507, the original image and the multiple result images may be directly input into the weight distribution model to obtain the modulation coefficient output by the weight distribution model.

In the above embodiment, a weight allocation model that can be used to allocate corresponding weight coefficients according to the noise reduction effect of the input image can be obtained by pre-training a neural network model. The modulation coefficient is obtained by inputting the result image into the weight allocation model, thereby improving the efficiency of image noise reduction processing.

FIG6 is a flow chart of an image noise reduction processing method in another embodiment of the present application. As shown in FIG6 , the process includes:

Step S270: input the original image into the image denoising model.

Step S280, obtaining an original denoised image corresponding to the original image output by the image denoising model.

Step S290: combining the original denoised image and the denoised image in a linear manner to obtain a target denoised image.

In some possible implementations, after obtaining a denoised image that is modulated by multiple result images, since the restored image before being processed by the image denoising model may lose some information in the original image during the scaling and quantization process, in order to avoid information omission, the original image can also be input into the image denoising model, and the denoising result corresponding to the original image can be linearly combined with the above-mentioned modulated denoised image to compensate for the information loss during the scaling and quantization process.

To further demonstrate the beneficial effects of the present application, a specific embodiment will be described below:

Assume that an image to be denoised is X, and the denoising model used to process the image to be denoised is F.

First, the image to be denoised can be normalized to compress the value range to [0, 1].

Assume that the scaling coefficient sequence C contains n scaling constants [C ₁ , C ₂ …C _n ], and for any scaling constant C _i , its value range is C _i > 1. Multiplying the image to be denoised by the scaling constant, we get the scaled image [y ₁ , y ₂ …y _n ], that is:

_yi ＝ _Ci · _Xi

Then, a quantization operation is applied to the scaled image to obtain a quantized image [z ₁ , z ₂ …z _n ], namely:

_zi = Quant( _yi )

Then, the quantized image is inversely scaled back to the original normalized value according to the scaling constant, and restored images of different frequencies [q ₁ , q ₂ …q _n ] can be obtained, that is:

To show how the frequency of the restored image changes compared to the image to be denoised, the change in the brightness value of the pixels in the image to be denoised is used as an example: suppose three pixels are selected in the image to be denoised for observation, and the brightness values of these three pixels are [0.11, 0.65, 0.11]. The brightness values are amplified to 40 times by the scaling constant, and the brightness values become [4.4, 26, 4.4]. The brightness values of the amplified image are rounded down to [4, 26, 4]. The brightness values are then inversely transformed according to the scaling constant to obtain the brightness values of the pixels in the restored image as [0.1, 0.65, 0.1]. It can be seen that compared with [0.11, 0.65, 0.11] in the image to be denoised, the high-frequency components in the restored image are higher, that is, a restored image with an overall higher frequency is obtained through scaling and quantization.

After obtaining the restored image, the restored image can be input into the denoising model F to obtain the output result image [p ₁ , p ₂ ..p _n ], and then the modulation coefficient is allocated according to the result image. In some embodiments, the allocation of the modulation coefficient can be achieved by manual intervention or using a neural network prediction.

Finally, the result images [p ₁ , p ₂ . . . p _n ] can be weighted combined according to the modulation coefficients to obtain the denoised image T.

In some other implementations, the image to be denoised may be input into the denoising model F to obtain the corresponding original denoising result image Tmp, and the denoised image T and the original denoising result image Tmp may be linearly combined to compensate for the quantization loss, thereby obtaining the target denoised image O, that is:

O＝αT+(1-α)Tmp

In the above formula, α is a constant greater than 0 and less than 1.

The denoising effect of the above embodiment can be seen in FIG7 , where the image on the left side of FIG7 is the image to be denoised, the image in the middle is the result of denoising the image to be denoised only by the denoising model, and the image on the right side is the result after modulation and linear combination according to the above method of this embodiment. As shown in FIG7 , since the denoising model performs denoising on a fixed frequency band, the overall high-frequency details of the denoised result (the image in the middle) are seriously lost, while after the method performs frequency division denoising and modulation combination, the high-frequency details in the rectangular area of the image on the right are significantly restored due to the reasonable frequency division denoising frequency band, and the overall denoising effect is better.

Corresponding to the above-mentioned embodiment of the image noise reduction processing method, the present application also provides an embodiment of an image noise reduction processing device 300. As shown in FIG8 , the device includes:

A scaling module 310 is used to scale the original image according to a scaling coefficient sequence to obtain a plurality of scaled images; the scaling coefficient sequence includes a plurality of constants;

A quantization module 320, configured to quantize the brightness values of target pixels in the scaled image to obtain a plurality of quantized images corresponding to the scaled image;

Restoration module 330, used for inversely scaling the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images;

An input module 340 is used to input a plurality of restored images into a preset image denoising model, and obtain a plurality of result images output by the image denoising model; the denoising model is a denoising model constructed based on a filter or a denoising model constructed based on a neural network;

An acquisition module 350 is used to determine a modulation coefficient of each result image according to a noise reduction effect of the result image;

The image modulation module 360 is used to weight and combine multiple result images according to the modulation coefficient to obtain a noise-reduced image.

For the specific definition of the image noise reduction processing device, please refer to the definition of the image noise reduction processing method above, which will not be repeated here. Each module in the above-mentioned image noise reduction processing device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

In an exemplary embodiment, the present application may utilize a computer device to implement the above-mentioned image noise reduction processing method.

In a possible implementation, the structure of the computer device is as shown in FIG. 9 . The computer device 100 includes at least one processor 110 , a memory 120 , a communication bus 130 , and at least one communication interface 140 .

The processor 110 may be a general-purpose central processing unit (CPU), a network processor (NP), a microprocessor, or may be one or more integrated circuits for implementing the solution of the present application, such as an application-specific integrated circuit (ASIC), a programmable logic device (PLD) or a combination thereof. The above-mentioned PLD may be a complex programmable logic device (CPLD), a field programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

The processor 110 may include one or more CPUs. The computer device 100 may include multiple processors 110. Each of these processors 110 may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). It should be noted that the processor 110 here may refer to one or more devices, circuits and/or processing cores for processing data (such as computer program instructions).

The memory 120 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, or a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. Optionally, the memory 120 may exist independently and be connected to the processor 110 via the communication bus 130; the memory 120 may also be integrated with the processor 110.

The communication bus 130 is used to transmit information between components (such as between a processor and a memory). The communication bus 120 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one communication bus is used in FIG1 , but it does not mean that there is only one bus or one type of bus. The communication interface 140 is used for the computer device 100 to communicate with other devices or communication networks.

The communication interface 140 includes a wired communication interface or a wireless communication interface. The wired communication interface may be, for example, an Ethernet interface. The Ethernet interface may be an optical interface, an electrical interface, or a combination thereof. The wireless communication interface may be a wireless local area network (WLAN) interface, a cellular network communication interface, or a combination thereof.

In some embodiments, the memory 120 can be used to store a computer program for executing the solution of the present application, and the processor 110 can execute the computer program stored in the memory 120. For example, the computer device 100 can call and execute the computer program stored in the memory 120 through the processor 110 to implement the steps of the image denoising processing method provided in the embodiment of the present application. It should be understood that the image denoising processing method provided in the present application can be applied to an image denoising processing device, which can be implemented as part or all of the processor 110 through software, hardware, or a combination of software and hardware, and integrated in the computer device 100.

In addition, it should be understood that the embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes a computer program. When the computer program is loaded and run on a computer device, the process or function shown in the embodiments of the present application is generated in whole or in part.

Among them, the computer program can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program can be transmitted from one website site, terminal, server or data center to another website site, terminal, server or data center via wired or wireless means.

The computer-readable storage medium may be any available medium that can be accessed by a computer device, or may be a data storage device such as a server or a data center that includes one or more available media.

It should be understood that the above is only a specific implementation of the embodiment of the present application, and is not intended to limit the protection scope of the embodiment of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solution of the embodiment of the present application shall be included in the protection scope of the embodiment of the present application.

Claims (10)

1. A method for image noise reduction, comprising: Scaling the original image according to a scaling coefficient sequence to obtain a plurality of scaled images; the scaling coefficient sequence includes a plurality of scaling constants; quantizing the brightness values of target pixels in the scaled image to obtain a plurality of quantized images corresponding to the scaled image; Inversely scale the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images; Inputting the multiple restored images into a preset image denoising model, and obtaining multiple result images output by the image denoising model; the image denoising model is a denoising model constructed based on a filter or a denoising model constructed based on a neural network; Determining a modulation coefficient of the result image according to a noise reduction effect of the result image; The multiple result images are weighted and combined according to the modulation coefficient to obtain a noise-reduced image. 2. The image denoising method according to claim 1, wherein scaling the original image according to the scaling coefficient sequence to obtain a plurality of scaled images comprises: Obtaining an image to be processed and a preset zoom factor sequence; Mapping the brightness value of each pixel in the image to be processed to a target interval to obtain an original image; The brightness values of the pixels in the original image are multiplied by the scaling constants in the scaling coefficient sequence to obtain a plurality of scaled images. 3. The image denoising method according to claim 2, wherein the target interval is [0, 1]. 4. The image denoising method according to claim 1 is characterized in that quantizing the brightness value of the target pixel in the scaled image includes performing one or more combinations of rounding down, rounding up, and rounding off on the brightness value of the target pixel. 5. The image denoising method according to claim 1, wherein determining the modulation coefficient of each result image according to the denoising effect of the result image comprises: Determining a denoising frequency band of the image denoising model according to the denoising effect of the result image; Calculating a noise frequency band of the result image according to brightness values of pixels in the result image, and counting overlapping frequency bands of the noise frequency band and the noise reduction frequency band; A modulation coefficient is assigned to the result image according to the ratio of the overlapping frequency band to the noise frequency band. 6. The image denoising method according to claim 1, wherein determining the modulation coefficient of the result image according to the denoising effect of the result image comprises: Calculating the signal-to-noise ratio of the result image; A modulation coefficient is assigned to the resulting image according to the signal-to-noise ratio. 7. The image denoising method according to claim 1, wherein determining the modulation coefficient of each result image according to the denoising effect of the result image comprises: Acquire a high-definition image of the original image; Inputting the high-definition image and the multiple result images into a pre-trained weight allocation model in sequence; the weight allocation model is a neural network model for allocating corresponding weight coefficients according to the noise reduction effect of the input image; Obtain a modulation coefficient output by the weight distribution model corresponding to the result image. 8. The image denoising method according to claim 1, characterized in that after the plurality of result images are weighted and combined according to the modulation coefficient to obtain the denoised image, the method further comprises: Inputting the original image into the image denoising model; Obtaining an original denoised image corresponding to the original image output by the image denoising model; The target denoised image is obtained by linearly combining the original denoised image and the denoised image. 9. An image noise reduction processing device, comprising: A scaling module, used for scaling the original image according to a scaling coefficient sequence to obtain a plurality of scaled images; the scaling coefficient sequence includes a plurality of constants; A quantization module, used to quantize the brightness value of the target pixel in the zoomed image to obtain a plurality of quantized images corresponding to the zoomed image; A restoration module, used for inversely scaling the quantized image according to the scaling coefficient sequence to obtain a plurality of restored images; An input module, used to input the multiple restored images into a preset image denoising model, and obtain multiple result images output by the image denoising model; the denoising model is a denoising model constructed based on a filter or a denoising model constructed based on a neural network; An acquisition module, used for determining a modulation coefficient of each result image according to a noise reduction effect of the result image; The image modulation module is used to weight and combine the multiple result images according to the modulation coefficient to obtain a noise-reduced image. 10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the method according to any one of claims 1 to 8 when executing the computer program.