Image processing method, device, equipment, system, storage medium and chip system

Abstract

The invention relates to an image processing method, device, equipment and system, a storage medium and a chip system, and the method comprises the steps: receiving a target noisy image and target semantic information, the target noisy image is an image received by a receiving end after a first image is transmitted through a wireless channel, the target semantic information is semantic information corresponding to the first image, and the first image and the target semantic information are sent to the receiving end by a sending end; and according to the target semantic information, performing denoising processing on the target noisy image through a pre-generated image processing model to obtain a second image, the second image having the same content as the first image. According to the method, the pre-generated image processing model is guided based on the target semantic information to denoise the target noisy image transmitted by the wireless channel, and the second image consistent with the first image in content is recovered, so that channel noise can be effectively resisted, and a high-quality and low-delay image transmission effect is obtained.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of communication, and in particular, to an image processing method, device, equipment, system, storage medium and chip system. BACKGROUND

[0002] In a wireless communication system, the channel transmission environment is complex and changeable, and the transmission information will be affected by channel noise. In a weak network environment, there may be a problem of data packet loss. If direct communication is performed in a non-retransmission manner, the received information will be incomplete. When retransmission mechanism is used to resend the lost data packet, the delay and bandwidth limitation of data transmission result in that the communication delay cannot be met. SUMMARY

[0003] To overcome the problems in the related art, the present disclosure provides an image processing method, device, equipment, system, storage medium and chip system.

[0004] According to a first aspect of an embodiment of the present disclosure, an image processing method is provided, executed by a receiving end, and the method comprises:

[0005] receiving a target noisy image and target semantic information, the target noisy image being an image received by the receiving end after a first image is transmitted through a wireless channel, the target semantic information being semantic information corresponding to the first image, and the first image and the target semantic information being transmitted by a sending end to the receiving end;

[0006] performing denoising processing on the target noisy image through a pre-generated image processing model according to the target semantic information, to obtain a second image, the second image having the same content as the first image.

[0007] Based on the target semantic information, the pre-generated image processing model is guided to perform denoising on the target noisy image after the wireless channel transmission, to restore a second image having the same content as the first image, which can effectively resist channel noise, thereby improving the effect of image transmission.

[0008] In some possible implementation manners, the performing denoising processing on the target noisy image through the pre-generated image processing model according to the target semantic information to obtain the second image comprises:

[0009] inputting the target noisy image, the target semantic information and a target noise image into the image processing model, to obtain the second image output by the image processing model, the target noise image being a randomly sampled noise image.

[0010] The target noisy image, target semantic information, and a randomly sampled target noise image are input into the image processing model for denoising processing, and a second image consistent with the content of the first image is effectively recovered.

[0011] In some possible implementations, the image processing model is pre-generated in the following manner:

[0012] A plurality of sample sets are obtained, the sample sets including a sample original image, a sample noisy image, and sample semantic information corresponding to the sample original image, the sample noisy image being an image obtained after the sample original image is transmitted through the wireless channel;

[0013] The target neural network model is trained through the plurality of sample sets, and the image processing model is obtained.

[0014] The target neural network model is trained through the plurality of sample sets, and the image processing model is obtained.

[0015] In some possible implementations, the target neural network model includes a first encoding sub-model, a decoding sub-model, and a second encoding sub-model having the same structure as the first encoding sub-model; and the training of the target neural network model through the plurality of sample sets to obtain the image processing model includes:

[0016] The model training step is cyclically performed according to the plurality of sample sets until the trained target neural network model meets a preset stop iteration condition, and the trained target neural network model is taken as the image processing model.

[0017] The model training step includes:

[0018] The sample noise image and the sample semantic information are input into the first encoding sub-model to obtain a plurality of first feature maps, the sample noise image being a randomly sampled noise image, and different first feature maps having different resolutions;

[0019] The sample noise image, the sample noisy image, and the sample semantic information are input into the second encoding sub-model to obtain a plurality of second feature maps, different second feature maps having different resolutions;

[0020] The plurality of first feature maps, the plurality of second feature maps, and the sample semantic information are input into the decoding sub-model to obtain a predicted denoised image.

[0021] A loss value is determined according to the predicted denoised image and the sample original image.

[0022] In a case where it is determined, according to the loss value, that the target neural network model does not satisfy the preset stop iteration condition, parameters of the second encoding sub-model are updated according to the loss value, to obtain a trained target neural network model, and the trained target neural network model is taken as a new target neural network model.

[0023] The second encoding sub-model can guide the training process of the image processing model by using the input sample noisy image, so that the image processing model can more accurately remove noise, thereby improving the effect of image denoising.

[0024] In some possible implementation manners, the first encoding sub-model includes a plurality of first encoding layers, different first encoding layers are used to output first feature maps of different resolutions, and the second encoding sub-model includes a plurality of second encoding layers, different second encoding layers are used to output second feature maps of different resolutions.

[0025] By setting a plurality of encoding layers in the first encoding sub-model and the second encoding sub-model to output feature maps of different resolutions, the image processing model can more comprehensively capture image information from a multi-scale feature level, and the accuracy of image denoising is further improved.

[0026] In some possible implementation manners, the decoding sub-model includes a plurality of decoding layers, different decoding layers correspond to different resolutions; and the inputting of the plurality of first feature maps, the plurality of second feature maps, and the sample semantic information into the decoding sub-model to obtain a predicted denoising image includes:

[0027] performing zero convolution processing on the second feature map to obtain a convolution feature map;

[0028] inputting the first feature map, the convolution feature map, and the sample semantic information into a decoding layer corresponding to a resolution to obtain the predicted denoising image.

[0029] By performing zero convolution processing on the first feature map and the second feature map respectively and then inputting the first feature map and the second feature map into a decoding layer corresponding to a resolution, multi-resolution feature fusion is implemented, so that the decoding sub-model can more accurately generate a predicted denoising image, and the accuracy of image denoising and recovery is improved.

[0030] In some possible implementation manners, the first image is an image obtained by downsampling an original image at the sending end, and the method further includes:

[0031] The image processing model is used to perform upsampling processing on the second image to obtain a third image, and the third image has the same size as the original image.

[0032] The third image with the same size as the original image can be obtained by upsampling the second image after downsampling and denoising, so as to realize accurate restoration of the image.

[0033] According to a second aspect of the embodiments of the present disclosure, an image processing method is provided, executed by a sending end, and the method comprises:

[0034] The sending end sends a first image and target semantic information to a receiving end, the target semantic information is semantic information corresponding to the first image, and the target semantic information is used for the receiving end to perform denoising on a target noisy image to obtain a second image, the target noisy image is an image received by the receiving end after the first image is transmitted through a wireless channel, and the second image has the same content as the first image.

[0035] In some possible implementation manners, the method further comprises:

[0036] The original image is subjected to downsampling to obtain the first image.

[0037] The sending end can transmit the first image subjected to downsampling to the receiving end, so as to reduce the transmission amount and improve the transmission efficiency.

[0038] In some possible implementation manners, the method further comprises:

[0039] The first image is input into a pre-generated semantic extraction model to obtain the target semantic information output by the semantic extraction model.

[0040] The target semantic information corresponding to the original image can be extracted through the semantic extraction model, so as to provide semantic support for image denoising based on semantic guidance.

[0041] According to a third aspect of the embodiments of the present disclosure, an image processing apparatus is provided, applied to a receiving end, and comprising:

[0042] A first transceiver module is configured to receive a target noisy image and target semantic information, the target noisy image is an image received by the receiving end after a first image is transmitted through a wireless channel, the target semantic information is semantic information corresponding to the first image, and the first image and the target semantic information are sent by a sending end to the receiving end.

[0043] A processing module is configured to perform denoising on the target noisy image through a pre-generated image processing model according to the target semantic information to obtain a second image, the second image has the same content as the first image.

[0044] According to a fourth aspect of the embodiments of the present disclosure, an image processing apparatus is provided, applied to a sending end, and comprising:

[0045] The second transceiving module is configured to send the first image and target semantic information to a receiving end, the target semantic information being semantic information corresponding to the first image, the target semantic information being used for the receiving end to perform denoising processing on a target noisy image to obtain a second image, the target noisy image being an image received by the receiving end after the first image is transmitted through a wireless channel, and the second image being the same in content as the first image.

[0046] According to a fifth aspect of an embodiment of the present disclosure, an image processing method is provided, including:

[0047] The sending end sends a first image and target semantic information to a receiving end, the target semantic information being semantic information corresponding to the first image;

[0048] The receiving end performs denoising processing on a target noisy image according to the target semantic information through a pre-generated image processing model to obtain a second image, the target noisy image being an image received by the receiving end after the first image is transmitted through a wireless channel, and the second image being the same in content as the first image.

[0049] According to a sixth aspect of an embodiment of the present disclosure, a communication device is provided, including:

[0050] One or more processors;

[0051] The processor is configured to call computer instructions to cause the communication device to perform the image processing method according to any one of the first aspect or the second aspect.

[0052] According to a seventh aspect of an embodiment of the present disclosure, an image processing system is provided, including a receiving end and a sending end, wherein the receiving end is configured to implement the image processing method according to any one of the first aspect, and the sending end is configured to implement the image processing method according to any one of the second aspect.

[0053] According to an eighth aspect of an embodiment of the present disclosure, a computer readable storage medium is provided, the computer readable storage medium storing computer instructions, when the computer instructions are run on a communication device, causing the communication device to perform the image processing method according to any one of the first aspect or the second aspect.

[0054] According to a ninth aspect of an embodiment of the present disclosure, a chip system is provided, including a processor and an interface, the processor being configured to read instructions to perform the image processing method according to any one of the first aspect or the second aspect.

[0055] According to a tenth aspect of the embodiments of the present disclosure, a computer program product is provided, which, when executed by a communication device, causes the communication device to perform the image processing method described in any one of the first aspect and the second aspect.

[0056] The technical solution provided by the embodiments of the present disclosure can have the following beneficial effects: a target noisy image and target semantic information are received, the target noisy image being an image received by a receiving end after a first image is transmitted through a wireless channel, the target semantic information being semantic information corresponding to the first image, the first image and the target semantic information being transmitted by a sending end to the receiving end; according to the target semantic information, a pre-generated image processing model is used to perform denoising processing on the target noisy image, to obtain a second image, the second image having the same content as the first image. The pre-generated image processing model is guided by the target semantic information to perform denoising processing on the target noisy image after the wireless channel transmission, to restore a second image having the same content as the first image, which can effectively resist channel noise, thereby obtaining a high-quality and low-delay image transmission effect.

[0057] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure together with the description.

[0059] Figure 1 is a semantic communication infrastructure diagram according to an exemplary embodiment.

[0060] Figure 2 is an interaction diagram of an image processing method according to an exemplary embodiment.

[0061] Figure 3 is a flowchart of a model training method according to an exemplary embodiment.

[0062] Figure 4 is a structural diagram of a target neural network model according to an exemplary embodiment.

[0063] Figure 5 is a flowchart of an image processing method according to an exemplary embodiment.

[0064] Figure 6 is a flowchart of another image processing method according to an exemplary embodiment.

[0065] Figure 7is a flowchart of another image processing method according to an example embodiment.

[0066] Figure 8 is a schematic diagram of image processing according to an example embodiment.

[0067] Figure 9 is a block diagram of an image processing apparatus according to an example embodiment.

[0068] Figure 10 is a block diagram of another image processing apparatus according to an example embodiment.

[0069] Figure 11 is a block diagram of a sending end according to an example embodiment.

[0070] Figure 12 is a block diagram of a receiving end according to an example embodiment.

[0071] Figure 13 is a schematic diagram of a chip system according to an example embodiment. DETAILED DESCRIPTION

[0072] The example embodiments will be described in detail herein with reference to the attached drawings. The following description is made with reference to the accompanying drawings in which like reference numerals refer to like elements, unless the context of use indicates otherwise. The following description of example embodiments is not representative of all embodiments consistent with the present disclosure. Rather, it is merely an example of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0073] It should be noted that all actions of acquiring signals, information or data in the present disclosure are performed in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization of the corresponding device owner.

[0074] Before introducing the image processing method, apparatus, device, system, storage medium and chip system provided by the present disclosure, the related technologies in the present scenario are first described. Semantic communication technology is a new wireless communication method, which generally uses schemes such as latent space transformation and joint source-channel coding (JSCC) to compress image data, and combines with technologies such as semantic importance coding to reduce communication loss, so as to improve information transmission performance.

[0075] Figure 1 is a semantic communication infrastructure diagram according to an example embodiment. As shown in Figure 1As shown, semantic communication is a task-based communication mode of "understanding first and then transmitting". It selectively extracts, compresses and transmits features of the original signal, and then communicates using semantic level information.

[0076] In some embodiments, semantic communication can be combined with deep learning to use latent space transformation to reduce the dimensionality of the original image to speed up the calculation. By joint source channel technology, semantic information is encoded at the transmitting end and the original signal is decoded at the receiving end. With the help of an entropy model, the signal to be transmitted is encoded in terms of semantic importance, increasing the noise resistance.

[0077] In some embodiments, semantic communication combines channel gain and adds Gaussian white noise to fit the wireless channel, and then trains the joint source channel coding based on the fitted channel. However, the actual channel transmission environment is more complex and variable than the channel estimation scheme under this ideal state, making it difficult for the current semantic communication scheme to resist actual channel noise in actual application. Especially when the communication is in a weak network environment, there is usually a problem of data packet loss. If communication is carried out directly in a non-retransmission manner, it will result in incomplete information received, affecting the accurate communication of semantics; and when a retransmission mechanism is used to resend the lost data packet, the delay and bandwidth limitation of data transmission will result in the inability to meet the communication delay.

[0078] To this end, the embodiments of the present disclosure provide an image processing method. The transmitting end extracts semantic information from the original image to be transmitted and converts it into text information to obtain target semantic information. Then the image and the target semantic information are transmitted through a wireless channel. The receiving end obtains a target noisy image with channel noise, inputs the target noisy image and the target semantic information into a trained image processing model, and generates a denoised image consistent with the original image information. This can effectively resist channel noise and obtain high-quality and low-delay image transmission results.

[0079] Figure 2 is a schematic diagram of an image processing method according to an example embodiment. As shown in Figure 2 The method includes:

[0080] In step S21, the transmitting end performs downsampling processing on the original image to obtain a first image.

[0081] In some embodiments, the original image can be an image that the transmitting end needs to send to the receiving end.

[0082] In some embodiments, when the transmitting end needs to send the original image to the receiving end, the original image can be first subjected to downsampling processing to obtain the first image. The first image can be understood as a thumbnail image of the original image.

[0083] In some embodiments, the resolution of the downsampling can be preconfigured. For example, the original image can be downsampling by 1 / 2 to obtain the first image.

[0084] In some embodiments, the resolution of the downsampling can be determined according to the current network state. The worse the network state, the higher the resolution of the downsampling, so as to obtain a smaller thumbnail image. In this way, the amount of data transmitted by the sending end can be controlled according to the network state.

[0085] In some embodiments, the sending end can also directly send the original image to the receiving end, that is, the first image is the original image, and in this case, step S21 can be omitted.

[0086] In step S22, the sending end inputs the first image into the pre-generated semantic extraction model to obtain target semantic information output by the semantic extraction model.

[0087] In some embodiments, the target semantic information is text information.

[0088] In some embodiments, the semantic extraction model can be a model for semantic extraction generated in related technologies, such as the BLIP-2 model, which will not be described here.

[0089] In some embodiments, the first image can be an original image or a thumbnail image obtained by downsampling the original image.

[0090] In some embodiments, the semantic extraction model can perform semantic extraction based on the first image and convert it into text information to obtain the target semantic information.

[0091] In step S23, the sending end sends the first image and the target semantic information to the receiving end.

[0092] In some embodiments, the first image and the target semantic information sent by the sending end to the receiving end, after passing through the wireless channel, the receiving end receives a first image with channel noise, that is, a target noisy image.

[0093] In step S24, the receiving end performs denoising processing on the target noisy image through a pre-generated image processing model according to the target semantic information to obtain a second image.

[0094] In some embodiments, the second image can be an image with the same content as the first image.

[0095] In some embodiments, after receiving the target noisy image and the target semantic information, the receiving end can randomly sample a latent noisy image as the input of the image processing model, input the target semantic information as the prompt input of the image processing model, and input the target noisy image as the conditional control input of the image processing model, and then perform denoising processing on the image processing model to obtain the second image.

[0096] In some embodiments, the target noisy image, the target semantic information, and the target noisy image are input into the image processing model to obtain the second image output by the image processing model.

[0097] In some embodiments, the target noisy image can be understood as the input of the image processing model, the target noisy image can be understood as the conditional control input of the image processing model, and the target semantic information can be understood as the prompt input of the image processing model.

[0098] In some embodiments, the target noisy image can be understood as the input of the image processing model, the target noisy image can be understood as the conditional control input of the image processing model, and the target semantic information can be understood as the prompt input of the image processing model.

[0099] In some embodiments, the image processing model can include a first encoding sub-model, a decoding sub-model, and a second encoding sub-model with the same structure as the first encoding sub-model. The input of the first encoding sub-model is the target noisy image and the target semantic information, the input of the second encoding sub-model is the target noisy image, the target semantic information, and the target noisy image, and the output of each layer of the first encoding sub-model and the second encoding sub-model can be used as the input of the corresponding layer of the decoding sub-model. The input of each layer of the decoding sub-model also includes the target semantic information, and the decoding sub-model finally outputs the second image.

[0100] In step S25, the receiving end performs up-sampling processing on the second image by using the image processing model to obtain a third image.

[0101] In some embodiments, if the first image is an image obtained by the sending end by down-sampling the original image, the receiving end can perform up-sampling processing on the second image to obtain a third image with the same size as the original image.

[0102] In some embodiments, the image processing model can also perform up-sampling processing. It should be understood that the image processing model can perform denoising processing and down-sampling processing to obtain a second image with the same content and size as the original image. In this case, step S25 can be omitted.

[0103] In some embodiments, the receiving end can also perform up-sampling processing on the second image after the image processing model outputs the second image.

[0104] In the technical solution, the sending end extracts semantic information from the original image to be transmitted and converts the semantic information into text information to obtain target semantic information, and then transmits the image and the target semantic information through a wireless channel. The receiving end obtains a target noisy image with channel noise, inputs the target noisy image and the target semantic information into a trained image processing model, and generates a denoised image consistent with the original image information. In this way, the receiving end can perform denoising processing in combination with the semantic information, effectively resist channel noise, and obtain high-quality and low-delay image transmission effect.

[0105] Figure 3 is a flowchart of a model training method according to an example embodiment. As shown in Figure 3 , the method comprises:

[0106] In step S31, a plurality of sample sets are obtained.

[0107] In some embodiments, the sample set can include a sample original image, a sample noisy image, and sample semantic information corresponding to the sample original image.

[0108] In some embodiments, the sample original image can be any image that can be collected.

[0109] In some embodiments, if the sending end transmits a thumbnail image to the receiving end, the collected image can be down-sampled to obtain a sample original image after thumbnailing. It should be understood that the sample original image has the same size as the first image.

[0110] In some embodiments, the sample noisy image is an image obtained after the sample original image is transmitted through a wireless channel. For example, the sample original image can be transmitted through an actual channel in various network environments to obtain a sample noisy image.

[0111] In some embodiments, the sample original image is subjected to semantic extraction through a semantic extraction model to obtain sample semantic information.

[0112] In some embodiments, the sample set can be a data pair in the format of <sample original image, sample noisy image, sample semantic information>.

[0113] In step S32, a target neural network model is trained through the plurality of sample sets to obtain an image processing model.

[0114] In some embodiments, the target neural network model includes a first encoding sub-model, a decoding sub-model, and a second encoding sub-model having the same structure as the first encoding sub-model.

[0115] In some embodiments, the model training step is performed cyclically according to the plurality of sample sets until the trained target neural network model meets a preset stop iteration condition, and the trained target neural network model is taken as the image processing model.

[0116] In some embodiments, when the model training step is performed for the first time, a sample set can be selected from the plurality of sample sets.

[0117] In some embodiments, the model training step includes:

[0118] Step S1, inputting a sample noise image and sample semantic information into a first encoding sub-model to obtain a plurality of first feature maps.

[0119] The sample noise image is a randomly sampled noise image, and the spatial resolutions of different first feature maps are different.

[0120] In some embodiments, the first encoding sub-model includes a plurality of first encoding layers, and different first encoding layers are used to output first feature maps with different resolutions.

[0121] Figure 4 is a structural schematic diagram of a target neural network model according to an exemplary embodiment. As shown in Figure 4 The input is a sample noise image, and the condition (input) is a sample noisy image. The first encoding sub-model includes five first encoding layers: SD Encoder Block_1 64*64, SD Encoder Block_2 32*32, SD Encoder Block_3 16*16, SD Encoder Block_4 8*8, and SD Middle Block 8*8.

[0122] The SD Encoder Block_1 64*64 can be understood as a feature extraction module in the first encoding sub-model, which is used to process the input sample original image to obtain a feature map with a spatial resolution of 64*64. Similarly, the first encoding sub-model can obtain first feature maps with spatial resolutions of 64*64, 32*32, 16*16, and 8*8, respectively.

[0123] In some embodiments, the input of each first encoding layer of the first encoding sub-model includes sample semantic information and temporal information. For example, the sample semantic information, temporal information, and a sample noise image are input into SD EncoderBlock_1 64*64 to obtain a first feature map with a spatial resolution of 64*64; the sample semantic information, temporal information, and a first feature map with a spatial resolution of 64*64 are input into SD Encoder Block_2 32*32 to obtain a first feature map with a spatial resolution of 32*32; the sample semantic information, temporal information, and a first feature map with a spatial resolution of 32*32 are input into SD Encoder Block_3 16*16 to obtain a first feature map with a spatial resolution of 16*16; and the sample semantic information, temporal information, and a first feature map with a spatial resolution of 16*16 are input into SD Encoder Block_48*8 to obtain a first feature map with a spatial resolution of 8*8.

[0124] Step S2: input the sample noisy image, the sample noisy image, and the sample semantic information into the second encoding sub-model to obtain a plurality of second feature maps.

[0125] Among them, different second feature maps have different resolutions.

[0126] In some embodiments, the second encoding sub-model includes multiple second encoding layers, and different second encoding layers are used to output second feature maps with different resolutions.

[0127] In some embodiments, referring to the first encoding sub-model of step S1, the second encoding sub-model can obtain second feature maps with spatial resolutions of 64*64, 32*32, 16*16, and 8*8 respectively.

[0128] In some embodiments, the parameters of the first encoding sub-model may be used as initial parameters of the second encoding sub-model.

[0129] Step S3: Input the plurality of first feature maps, the plurality of second feature maps, and the sample semantic information into the decoding sub-model to obtain a predicted denoised image.

[0130] In some embodiments, the decoding sub-model includes multiple decoding layers, and different decoding layers correspond to different resolutions. Figure 4As shown, the decoding sub-model can include four decoding layers: SD Decoder Block_4 8*8, SD Decoder Block_3 16*16, SD Decoder Block_2 32*32, and SD Decoder Block_1 64*64.

[0131] In some embodiments, the sample semantic information, the time information, and the first feature map and the second feature map with a spatial resolution of 8*8 are input into the SD Decoder Block_4 8*8 to obtain a third feature map with a spatial resolution of 8*8; the sample semantic information, the time information, and the first feature map and the second feature map with a spatial resolution of 16*16 are input into the SD Decoder Block_3 16*16 to obtain a third feature map with a spatial resolution of 16*16; the sample semantic information, the time information, and the first feature map and the second feature map with a spatial resolution of 32*32 are input into the SD Decoder Block_2 32*32 to obtain a third feature map with a spatial resolution of 32*32; the sample semantic information, the time information, and the first feature map and the second feature map with a spatial resolution of 64*64 are input into the SD Decoder Block_2 64*64 to obtain a third feature map with a spatial resolution of 64*64; and finally, the third feature map with a spatial resolution of 64*64 is restored by the VAE decoder (not shown in the VAE decoder) to obtain the predicted denoised image. Figure 4

[0132] In some embodiments, the second feature map is subjected to zero convolution processing to obtain a convolution feature map; the first feature map, the convolution feature map, and the sample semantic information are input into a decoding layer corresponding to the resolution to obtain the predicted denoised image.

[0133] In some embodiments, the second feature map with a spatial resolution of 64*64 is subjected to zero convolution processing to obtain a convolution feature map with a spatial resolution of 64*64, the second feature map with a spatial resolution of 32*32 is subjected to zero convolution processing to obtain a convolution feature map with a spatial resolution of 32*32, the second feature map with a spatial resolution of 16*16 is subjected to zero convolution processing to obtain a convolution feature map with a spatial resolution of 16*16, and the second feature map with a spatial resolution of 8*8 is subjected to zero convolution processing to obtain a convolution feature map with a spatial resolution of 8*8.

[0134] ​In some embodiments, the output of the SD Middle Block 8*8 in the second encoding submodel can be fused with the output of the SD Middle Block 8*8 in the first encoding submodel, and the fusion result is taken as the input of the SD Decoder Block_4 8*8 in the decoding submodel. Meanwhile, the input of the SD Decoder Block_4 8*8 also includes the first feature map with a spatial resolution of 8*8 and the convolution feature map. Similarly, the input of the SD Decoder Block_3 16*16 also includes the first feature map with a spatial resolution of 16*16 and the convolution feature map, the input of the SD Decoder Block_2 32*32 also includes the first feature map with a spatial resolution of 32*32 and the convolution feature map, and the input of the SD Decoder Block_1 64*64 also includes the first feature map with a spatial resolution of 64*64 and the convolution feature map. Finally, the output of the SD Decoder Block_1 64*64 is restored by the VAE decoder (not shown in the figure) to obtain the predicted denoised image. Figure 4

[0135] Step S4: determining a loss value according to the predicted denoised image and the sample original image.

[0136] In some embodiments, the predicted denoised image and the sample original image are the same in content and size.

[0137] In some embodiments, the loss value is determined based on a reconstruction loss according to the predicted denoised image and the sample original image.

[0138] Step S5: in the case where the target neural network model does not satisfy the preset stop iteration condition according to the loss value, updating the parameters of the second encoding submodel according to the loss value to obtain a trained target neural network model, and taking the trained target neural network model as a new target neural network model.

[0139] In some embodiments, the preset stop iteration condition can be any stop iteration condition in the prior art, for example, the loss value is less than or equal to a preset loss value threshold, and the present disclosure does not make any limitation in this regard.

[0140] ​In some embodiments, after determining the loss value, it can be determined whether the loss value is less than or equal to a preset loss value threshold. If the loss value is less than or equal to the preset loss value threshold, it can be determined that the target neural network model meets the preset stopping iteration condition. If the loss value is greater than the preset loss value threshold, it can be determined that the target neural network model does not meet the preset stopping iteration condition, the parameters of the second encoding sub-model are updated according to the loss value, a trained target neural network model is obtained, the trained target neural network model is used as a new target neural network model, a new sample set is determined from the plurality of sample sets, and steps S1-S4 are continued to be performed according to the new sample set.

[0141] In some embodiments, the loss value is also used to update the weight value of the zero convolution processing.

[0142] In some embodiments, if the loss value is less than or equal to the preset loss value threshold, it can be determined that the target neural network model meets the preset stopping iteration condition, and the target neural network model is used as the image processing model. It should be understood that the image processing model has the same structure as the target neural network model.

[0143] Figure 5 is a flowchart of an image processing method according to an exemplary embodiment. As shown in Figure 5 , the method can be performed by a receiving end. The method can include:

[0144] In step S51, a target noisy image and target semantic information are received.

[0145] The optional implementation of step S51 can refer to the optional implementation of step S23 of Figure 2 and other related parts of the embodiments involved, which will not be repeated here. Figure 2

[0146] In step S52, according to the target semantic information, the target noisy image is denoised by the pre-generated image processing model to obtain a second image.

[0147] The optional implementation of step S52 can refer to the optional implementation of step S24 of Figure 2 and other related parts of the embodiments involved, which will not be repeated here. Figure 2 In some embodiments, reference can be made to each step and its optional implementation in other embodiments described before or after the description of the present embodiment, and other related parts of the description, which will not be repeated here.

[0148]

[0149] Figure 6 ​​FIG. 1 is a flow chart of another image processing method according to an exemplary embodiment. Figure 6 As shown, the method can be performed by the sending end. The method may include:

[0150] In step S61, downsampling is performed on the original image to obtain a first image.

[0151] The optional implementation of step S61 can be found in Figure 2 Optional implementation of step S21, and Figure 2 Other related parts in the embodiments involved will not be described in detail here.

[0152] In step S62 , the first image is input into a pre-generated semantic extraction model to obtain target semantic information output by the semantic extraction model.

[0153] The optional implementation of step S62 can be found in Figure 2 Optional implementation of step S22, and Figure 2 Other related parts in the embodiments involved will not be described in detail here.

[0154] In step S63, the first image and target semantic information are sent.

[0155] The optional implementation of step S63 can be found in Figure 2 Optional implementation of step S23, and Figure 2 Other related parts in the embodiments involved will not be described in detail here.

[0156] In some embodiments, reference may be made to the steps and optional implementation methods of other embodiments recorded before or after the description corresponding to this embodiment, as well as other related parts in the description, which will not be repeated here.

[0157] Figure 7 FIG. 1 is a flow chart of another image processing method according to an exemplary embodiment. Figure 7 As shown, the method can be performed by the sending end. The method may include:

[0158] In step S71 , a first image and target semantic information are sent.

[0159] The optional implementation of step S71 can be found in Figure 2 Optional implementation of step S23, and Figure 2 Other related parts in the embodiments involved will not be described in detail here.

[0160] In some embodiments, reference may be made to the steps and optional implementation methods of other embodiments recorded before or after the description corresponding to this embodiment, as well as other related parts in the description, which will not be repeated here.

[0161] Figure 8 FIG. 1 is a schematic diagram showing an image processing according to an exemplary embodiment. Figure 8 As shown, the sending end adjusts the size of the original image to obtain a thumbnail image, performs semantic extraction on the thumbnail image, generates semantic text information, and transmits the thumbnail image and semantic text information through a wireless channel. The receiving end receives the noisy image and semantic text information, and performs denoising processing through the image processing model to obtain a denoised image.

[0162] Figure 9 FIG. 1 is a block diagram of an image processing apparatus according to an exemplary embodiment. Figure 9 The image processing device 900 includes a first transceiver module 910 and a processing module 920.

[0163] The first transceiver module 910 is configured to receive a target noisy image and target semantic information, wherein the target noisy image is an image received by the receiving end after the first image is transmitted through a wireless channel, and the target semantic information is semantic information corresponding to the first image, and the first image and the target semantic information are sent by the transmitting end to the receiving end;

[0164] The processing module 920 is configured to perform denoising on the target noisy image according to the target semantic information using a pre-generated image processing model to obtain a second image, where the second image has the same content as the first image.

[0165] Optionally, the processing module 920 is further configured to input the target noisy image, the target semantic information and the target noisy image into the image processing model to obtain the second image output by the image processing model, where the target noisy image is a randomly sampled noise image.

[0166] Optionally, the processing module 920 is further configured to:

[0167] Acquire multiple sample sets, the sample sets including sample original images, sample noisy images, and sample semantic information corresponding to the sample original images, wherein the sample noisy images are images obtained after the sample original images are transmitted through the wireless channel;

[0168] The target neural network model is trained using a plurality of the sample sets to obtain the image processing model.

[0169] Optionally, the target neural network model includes a first encoding sub-model, a decoding sub-model, and a second encoding sub-model having the same structure as the first encoding sub-model; the processing module 920 is further configured to:

[0170] The model training step is performed according to a plurality of sample set cycles until the trained target neural network model meets a preset stop iteration condition, and the trained target neural network model is taken as the image processing model.

[0171] The model training step includes:

[0172] The sample noise image and the sample semantic information are input into the first encoding sub-model to obtain a plurality of first feature maps, the sample noise image is a randomly sampled noise image, and different first feature maps have different resolutions;

[0173] The sample noise image, the sample noise image, and the sample semantic information are input into the second encoding sub-model to obtain a plurality of second feature maps, and different second feature maps have different resolutions;

[0174] The plurality of first feature maps, the plurality of second feature maps, and the sample semantic information are input into the decoding sub-model to obtain a predicted denoising image;

[0175] According to the predicted denoising image and the sample original image, a loss value is determined;

[0176] In a case where it is determined according to the loss value that the target neural network model does not meet the preset stop iteration condition, parameters of the second encoding sub-model are updated according to the loss value to obtain a trained target neural network model, and the trained target neural network model is taken as a new target neural network model.

[0177] Optionally, the first encoding sub-model includes a plurality of first encoding layers, different first encoding layers are used to output first feature maps with different resolutions, and the second encoding sub-model includes a plurality of second encoding layers, different second encoding layers are used to output second feature maps with different resolutions.

[0178] Optionally, the decoding sub-model includes a plurality of decoding layers, different decoding layers correspond to different resolutions; the processing module 920 is further configured to:

[0179] The second feature map is subjected to zero convolution processing to obtain a convolution feature map;

[0180] The first feature map, the convolution feature map, and the sample semantic information are input into a decoding layer corresponding to a resolution to obtain the predicted denoising image.

[0181] Optionally, the first image is an image obtained by downsampling an original image by the sending end, and the processing module 920 is further configured to perform upsampling processing on the second image by the image processing model to obtain a third image, the third image has the same size as the original image.

[0182] Figure 10 is a block diagram of another image processing apparatus according to an exemplary embodiment. Referring to Figure 10 , the image processing apparatus 1000 includes a second transceiving module 1010.

[0183] The second transceiving module 1010 is configured to send a first image and target semantic information to a receiving end, the target semantic information being semantic information corresponding to the first image, the target semantic information being used by the receiving end to perform denoising processing on a target noisy image to obtain a second image, the target noisy image being an image received by the receiving end after the first image is transmitted through a wireless channel, the second image having the same content as the first image.

[0184] Optionally, the image processing apparatus 1000 further includes a second processing module configured to perform down-sampling processing on an original image to obtain the first image.

[0185] Optionally, the second processing module is further configured to input the first image into a pre-generated semantic extraction model to obtain the target semantic information output by the semantic extraction model.

[0186] As to the apparatus in the above-described embodiments, specific manners in which various modules perform operations have been described in details in the embodiments of the method, and thus will not be described in details here.

[0187] The present disclosure also provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, implement the steps of the image processing method provided by the present disclosure.

[0188] Figure 11 is a block diagram of a sending end according to an exemplary embodiment. For example, the sending end 1100 can be a mobile phone, a computer, a digital broadcast terminal, a messaging equipment, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0189] Referring to Figure 11 , the sending end 1100 can include one or more of the following components: a first processing component 1102, a first memory 1104, a first power supply component 1106, a multimedia component 1108, an audio component 1110, a first input / output interface 1112, a sensor component 1114, and a communication component 1116.

[0190] The first processing component 1102 generally controls the overall operations of the sending end 1100, such as operations associated with displays, phone calls, data communications, camera operations, and recording operations. The first processing component 1102 can include one or more first processors 1120 to execute instructions to complete all or part of the steps of the image processing method described above. In addition, the first processing component 1102 can include one or more modules to facilitate the interaction between the first processing component 1102 and other components. For example, the first processing component 1102 can include a multimedia module to facilitate the interaction between the multimedia component 1108 and the first processing component 1102.

[0191] The first memory 1104 is configured to store various types of data to support the operations of the sending end 1100. Examples of these data include instructions for any application or method operating on the sending end 1100, contact data, phonebook data, messages, pictures, videos, and the like. The first memory 1104 can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0192] The first power component 1106 provides power to the various components of the sending end 1100. The first power component 1106 can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the sending end 1100.

[0193] The multimedia component 1108 includes a screen that provides an output interface between the sending end 1100 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, and a gesture on the touch panel. The touch sensors can not only sense a boundary of a touch or a slide action, but also detect duration and pressure related to the touch or slide action. In some embodiments, the multimedia component 1108 includes a front camera and / or a rear camera. The front camera and / or the rear camera can receive external multimedia data when the sending end 1100 is in an operation mode, such as a photographing mode or a video mode. Each of the front camera and the rear camera can be a fixed optical lens system or have a focal length and optical zoom capability.

[0194] The audio component 1110 is configured to output and / or input audio signals. For example, the audio component 1110 includes a microphone (MIC) that is configured to receive an external audio signal when the sending end 1100 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the first memory 1104 or transmitted via the communication component 1116. In some embodiments, the audio component 1110 also includes a speaker for outputting audio signals.

[0195] The first input / output interface 1112 provides an interface between the first processing component 1102 and peripheral interface modules, which can be a keypad, a click wheel, buttons, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0196] The sensor component 1114 includes one or more sensors for providing status assessments of various aspects of the sending end 1100. For example, the sensor component 1114 can detect an open / closed state of the sending end 1100, relative positioning of components, such as a display and a keypad of the sending end 1100, a change in position of the sending end 1100 or a component of the sending end 1100, presence or absence of user contact with the sending end 1100, orientation or acceleration / deceleration / g-force and temperature of the sending end 1100. The sensor component 1114 can include an orientation sensor, a proximity sensor configured to detect presence of an object in a proximity without any physical touch, or a motion sensor configured to detect motion of the sending end 1100. The sensor component 1114 can further include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor component 1114 can also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0197] The communication component 1116 is configured to facilitate wired or wireless communication between the sending end 1100 and other devices. The sending end 1100 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an example embodiment, the communication component 1116 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In an example embodiment, the communication component 1116 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technology.

[0198] In exemplary embodiments, the sending end 1100 can be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components, for executing the above-mentioned image processing method.

[0199] In exemplary embodiments, a non-transitory computer readable storage medium including instructions, such as the first memory 1104 including instructions, is also provided, which can be executed by the first processor 1120 of the sending end 1100 to complete the above-mentioned image processing method. For example, the non-transitory computer readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0200] In another exemplary embodiment, a computer program product is also provided, which contains a computer program executable by a programmable apparatus, the computer program having code portions for performing the above-mentioned image processing method when executed by the programmable apparatus.

[0201] Figure 12 is a block diagram of a receiving end according to an exemplary embodiment. For example, the receiving end 1200 can be provided as a server. Referring to Figure 12 , the receiving end 1200 includes a second processing component 1222, which further includes one or more processors, and memory resources represented by a second memory 1232 for storing instructions executable by the second processing component 1222, such as an application program. The application program stored in the second memory 1232 can include one or more than one module each corresponding to a set of instructions. In addition, the second processing component 1222 is configured to execute instructions to perform the above-mentioned image processing method.

[0202] The receiving end 1200 can also include a second power supply component 1226 configured to perform power management of the receiving end 1200, a wired or wireless network interface 1250 configured to connect the receiving end 1200 to a network, and a second input / output interface 1258. The receiving end 1200 can operate based on an operating system stored in the memory 1232, such as Windows Server TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TM or the like.

[0203] Some embodiments of the present disclosure also provide a chip system, such as Figure 13As shown, the chip system includes at least one second processor 1301 and at least one interface circuit 1302. The second processor 1301 and the interface circuit 1302 can be interconnected by a line. For example, the interface circuit 1302 can be used to receive signals from other devices (e.g., a memory of an electronic device). For another example, the interface circuit 1302 can be used to send signals to other devices (e.g., the second processor 1301). Illustratively, the interface circuit 1302 can read instructions stored in a memory and send the instructions to the second processor 1301. When the instructions are executed by the processor 1301, the image processing device can perform various steps in the above-described embodiments. Of course, the chip system can also include other discrete devices, which are not specifically limited in some embodiments of the present disclosure.

[0204] In some embodiments of the present disclosure, the interface circuit 1302 can obtain data, program instructions and / or information, etc. in an internal storage area of the chip system; or can obtain data, program instructions and / or information, etc. outside the chip system.

[0205] Optionally, the chip system further includes a third memory 1303, which is used to store necessary computer programs and data.

[0206] Those skilled in the art can understand that the various illustrative logical blocks (illustrative logical blocks) and steps (steps) listed in the embodiments of the present application can be implemented by electronic hardware, computer software, or a combination of both. Whether the function is implemented by hardware or software depends on the specific application and design requirements of the whole system. Those skilled in the art can use various methods to implement the functions for each specific application, but such implementation should not be understood as beyond the scope of protection of the embodiments of the present application.

[0207] Although terms such as "first", "second" and "third" can be used herein to describe various components, parts, regions, layers or sections, these components, parts, regions, layers or sections are not limited by these terms. Instead, these terms are only used to distinguish one component, part, region, layer or section from another component, part, region, layer or section. Therefore, the first component, part, region, layer or section mentioned in the examples described herein can also be referred to as the second component, part, region, layer or section without departing from the teachings of the examples. In addition, the terms "first", "second" are only for description purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description herein, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0208] Also, the word "example" is used herein to mean serving as an example, instance, or illustration. Any implementation described herein as "example" is not necessarily to be construed as preferred or advantageous over other implementations. Rather, use of the word "example" is intended to present concepts in a concrete fashion. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specified otherwise, or clear from context, "X employs A or B" is intended to mean any of the natural inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied under any of the foregoing instances. In addition, the articles "a" and "an" as used in this application and the appended claims should generally be construed to mean "one or more" unless specified otherwise or clear from context to be directed to a singular form. Thus, use of the articles in this application and the following claims is not limiting.

[0209] Also, although the disclosure has been described in relation to one implementation, many options modifications and variations can occur to those skilled in the art upon reading this disclosure. The disclosure is exemplary, and the intended scope of the disclosure is not to be addressed by the scope of the claims. In particular, with respect to the various functions described above with regard to the components (e.g., elements, resources, etc.) described above, unless otherwise indicated, the terminology used to describe such components is intended to correspond to any component that performs the particular function of the described component (functionally equivalent), even if structurally not equivalent to the disclosed structure. In addition, although a particular feature of the disclosure can have been disclosed with respect to only one of several implementations, such feature can be combined with one or more other features of the other implementations as can be desired and advantageous for any given or particular application. Furthermore, to the extent that "comprising", "having", "containing", "including", or the like, are used in the detailed description or the claims, such terms are used in the sense of a

[0210] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the features disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure that come within known use or custom in the art to which the disclosure pertains. The specification and examples are to be regarded as illustrative only, and the true scope and spirit of the disclosure are indicated by the appended claims.

[0211] It should be understood that the present disclosure is not limited to the precise construction that has been described above and shown in the accompanying drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the present disclosure. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An image processing method, characterized in that: Executed by the receiving end, the method includes: receiving a target noisy image and target semantic information, wherein the target noisy image is an image received by the receiving end after the first image is transmitted through a wireless channel, and the target semantic information is semantic information corresponding to the first image, and the first image and the target semantic information are sent by the transmitting end to the receiving end; According to the target semantic information, the target noisy image is denoised using a pre-generated image processing model to obtain a second image, where the second image has the same content as the first image.

2. The method according to claim 1, characterized in that The step of performing denoising on the target noisy image using a pre-generated image processing model according to the target semantic information to obtain a second image includes: The target noisy image, the target semantic information and the target noisy image are input into the image processing model to obtain the second image output by the image processing model, where the target noisy image is a randomly sampled noise image.

3. The method according to claim 1 or 2, characterized in that The image processing model is pre-generated in the following way: Acquire multiple sample sets, the sample sets including sample original images, sample noisy images, and sample semantic information corresponding to the sample original images, wherein the sample noisy images are images obtained after the sample original images are transmitted through the wireless channel; The target neural network model is trained using a plurality of the sample sets to obtain the image processing model.

4. The method according to claim 3, characterized in that The target neural network model includes a first encoding sub-model, a decoding sub-model, and a second encoding sub-model with the same structure as the first encoding sub-model; the target neural network model is trained by using a plurality of sample sets to obtain the image processing model, including: cyclically executing a model training step according to the plurality of sample sets until the trained target neural network model satisfies a preset stop iteration condition, and using the trained target neural network model as the image processing model; The model training steps include: Inputting a sample noise image and the sample semantic information into the first encoding sub-model to obtain a plurality of first feature maps, wherein the sample noise image is a randomly sampled noise image, and different first feature maps have different resolutions; Inputting the sample noisy image, the sample noisy image, and the sample semantic information into the second encoding sub-model to obtain a plurality of second feature maps, where different second feature maps have different resolutions; Inputting a plurality of the first feature maps, a plurality of the second feature maps, and the sample semantic information into the decoding sub-model to obtain a predicted denoised image; Determining a loss value according to the predicted denoised image and the sample original image; When it is determined based on the loss value that the target neural network model does not meet the preset stop iteration condition, the parameters of the second encoding sub-model are updated based on the loss value to obtain the trained target neural network model, and the trained target neural network model is used as the new target neural network model.

5. The method according to claim 4, characterized in that The first encoding sub-model includes multiple first encoding layers, and different first encoding layers are used to output first feature maps with different resolutions. The second encoding sub-model includes multiple second encoding layers, and different second encoding layers are used to output second feature maps with different resolutions.

6. The method according to claim 4 or 5, characterized in that The decoding sub-model includes multiple decoding layers, and different decoding layers correspond to different resolutions; Inputting the plurality of first feature maps, the plurality of second feature maps, and the sample semantic information into the decoding sub-model to obtain a predicted denoised image includes: Performing zero convolution on the second feature map to obtain a convolution feature map; The first feature map, the convolution feature map, and the sample semantic information are input into a decoding layer of corresponding resolution to obtain the predicted denoised image.

7. The method according to claim 1, characterized in that The first image is an image obtained by downsampling the original image by the transmitting end, and the method further includes: The second image is up-sampled by the image processing model to obtain a third image, where the third image has the same size as the original image.

8. An image processing method, characterized in that: Executed by the sending end, the method includes: A first image and target semantic information are sent to a receiving end, where the target semantic information is semantic information corresponding to the first image, and the target semantic information is used by the receiving end to denoise the target noisy image to obtain a second image. The target noisy image is an image received by the receiving end after the first image is transmitted through a wireless channel, and the second image has the same content as the first image.

9. The method according to claim 8, characterized in that The method further comprises: Downsampling is performed on the original image to obtain the first image.

10. The method according to claim 8 or 9, characterized in that The method further comprises: The first image is input into a pre-generated semantic extraction model to obtain the target semantic information output by the semantic extraction model.

11. An image processing device, characterized in that: Applied to the receiving end, including: a first transceiver module configured to receive a target noisy image and target semantic information, wherein the target noisy image is an image received by the receiving end after the first image is transmitted via a wireless channel, and the target semantic information is semantic information corresponding to the first image, and the first image and the target semantic information are sent by the transmitting end to the receiving end; The processing module is configured to perform denoising on the target noisy image according to the target semantic information using a pre-generated image processing model to obtain a second image, where the content of the second image is the same as that of the first image.

12. An image processing device, characterized in that: Applied to the sending end, including: The second transceiver module is configured to send a first image and target semantic information to a receiving end, where the target semantic information is semantic information corresponding to the first image. The target semantic information is used by the receiving end to denoise the target noisy image to obtain a second image. The target noisy image is an image received by the receiving end after the first image is transmitted through a wireless channel. The second image has the same content as the first image.

13. An image processing method, characterized in that: include: The transmitting end sends the first image and target semantic information to the receiving end, where the target semantic information is semantic information corresponding to the first image; The receiving end performs denoising on the target noisy image using a pre-generated image processing model based on the target semantic information to obtain a second image, where the target noisy image is the image received by the receiving end after the first image is transmitted through a wireless channel, and the content of the second image is the same as that of the first image.

14. A communication device, characterized in that: include: one or more processors; The processor is used to call computer instructions so that the communication device executes the image processing method according to any one of claims 1 to 7; or the processor is used to call computer instructions so that the communication device executes the image processing method according to any one of claims 8 to 10.

15. An image processing system, characterized in that: The invention comprises a receiving end and a transmitting end, wherein the receiving end is configured to implement the image processing method according to any one of claims 1 to 7, and the transmitting end is configured to implement the image processing method according to any one of claims 8 to 10.

16. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented; or, when the computer program instructions are executed by a processor, the steps of the method according to any one of claims 8 to 10 are implemented.

17. A chip system, characterized in that: The method comprises a processor and an interface, wherein the processor is used to read instructions to execute the method according to any one of claims 1 to 7; or the processor is used to read instructions to execute the method according to any one of claims 8 to 10.

Images