Image quality detection method and device and storage medium

Abstract

The invention provides an image quality detection method and device and a storage medium, and the method comprises the steps: extracting the image features of a to-be-detected image through a feature extraction network, and carrying out the recognition of the image features through a global anomaly detection network and a local anomaly detection network; therefore, parameters representing the fuzzy degree of global blurring of the to-be-detected image and a local quality abnormal area of the to-be-detected image are obtained. Compared with the traditional image quality diagnosis which depends on the selection of manual evaluation indexes, the image quality is directly evaluated by using the deep neural network, the subjectivity of artificial features is avoided, and the accuracy of quality detection is improved.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to an image quality detection method, device and storage medium. Background Art

[0002] In autonomous driving scenarios, image quality affects the perception and decision-making process, so it is particularly important to accurately identify image quality issues.

[0003] Traditional image quality diagnosis comprehensively evaluates the quality of images through manually defined evaluation indicators (such as signal-to-noise ratio, white balance, color difference and distortion, etc.). This method relies too much on manually defined evaluation indicators and is highly subjective. In addition, these manually defined evaluation indicators are usually not suitable for the application scenario of autonomous driving. Summary of the invention

[0004] The present application proposes an image quality detection method, device and storage medium, which can alleviate the technical problem in the related art that the detection accuracy is not high due to the mismatch between the receptive field and the target object size.

[0005] In a first aspect, an embodiment of the present application proposes an image quality detection method, which is applied to an image anomaly detection network, wherein the image anomaly detection network includes a feature extraction network, a global anomaly detection network, and a local anomaly detection network, wherein the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; the method includes:

[0006] Extracting image features of the image to be detected through the feature extraction network;

[0007] The global anomaly detection network is used to identify the image features to obtain parameters characterizing the degree of blur of the global blur of the image to be detected; and a local anomaly detection network is used to obtain regional parameters characterizing the local quality abnormality area in the image to be detected based on the image features, and the local quality abnormality area is marked on the image to be detected according to the regional parameters.

[0008] In some embodiments, the feature extraction network includes a semantic information extraction module and a texture information extraction module, the image features include semantic information features extracted by the semantic information extraction module and texture information features extracted by the texture information extraction module; the local anomaly detection network includes a fusion module and a feature recognition module;

[0009] Using a local anomaly detection network and based on the image features, obtaining regional parameters characterizing a local quality anomaly region in the image to be detected, including:

[0010] The semantic information feature and the texture information feature are fused by the fusion module to obtain a fusion feature;

[0011] The feature recognition module is used to recognize the fusion feature to obtain the region parameter.

[0012] In some embodiments, the feature recognition module includes a first network, a second network, and a third network connected in parallel; using the feature recognition module to recognize the fusion feature to obtain the region parameter includes:

[0013] The first network is used to identify the fused features to obtain a position parameter representing the position of the local quality abnormality region in the image to be detected; the second network is used to identify the fused features to obtain a type parameter representing the abnormality type of the local quality abnormality region; and the third network is used to identify the fused features to obtain a size parameter representing the size of the local quality abnormality region;

[0014] The position parameter, the type parameter, and the size parameter are determined as the region parameters.

[0015] In some embodiments, the first network, the second network, and the third network each include a plurality of convolutional layers connected in series.

[0016] In some embodiments, the feature extraction network includes a convolutional layer and multiple residual convolutional layers connected to the convolutional layer, and the multiple residual convolutional layers are connected in series; the semantic information extraction module includes a first residual convolutional layer, and the texture information extraction module includes a last residual convolutional layer, the first residual convolutional layer is the first to be connected to the convolutional layer among the multiple residual convolutional layers, and the last residual convolutional layer is the last to be connected to the convolutional layer among the multiple residual convolutional layers.

[0017] In some embodiments, the fusion module includes a preset data connection function;

[0018] The fusion module is used to fuse the semantic information feature and the texture information feature to obtain a fusion feature, including:

[0019] The data connection function is called to concatenate the semantic information feature and the texture information feature to obtain the fusion feature.

[0020] In some embodiments, the global anomaly detection network includes a pooling layer, a fully connected layer, and an activation function layer;

[0021] The pooling layer is connected to the feature extraction network and the fully connected layer respectively;

[0022] The fully connected layer is also connected to the activation function layer.

[0023] In a second aspect, an embodiment of the present application proposes an image quality detection device, which is applied to an image anomaly detection network, wherein the image anomaly detection network includes a feature extraction network, a global anomaly detection network, and a local anomaly detection network, wherein the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; the device includes:

[0024] An extraction module, used for extracting image features of the image to be detected through the feature extraction network;

[0025] A processing module is used to use the global anomaly detection network to identify the image features and obtain parameters that characterize the degree of blur of the global blur of the image to be detected; and, use a local anomaly detection network and, based on the image features, obtain regional parameters that characterize the local quality abnormality area in the image to be detected, and mark the local quality abnormality area on the image to be detected according to the regional parameters.

[0026] A third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect.

[0027] A fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the method described in the first aspect.

[0028] The technical solution provided in the embodiments of the present application has at least the following technical effects or advantages:

[0029] In the embodiment of the present application, the image features of the image to be detected are extracted by a feature extraction network, and the global anomaly detection network and the local anomaly detection network are used to identify the image features respectively, so as to obtain the parameters representing the degree of blur of the global blur of the image to be detected and the local quality abnormal area of ​​the image to be detected. Compared with the traditional image quality diagnosis that relies on the selection of manual evaluation indicators, the present application uses a deep neural network to directly evaluate the quality of the image, avoiding the subjectivity of manual features and improving the accuracy of quality detection.

[0030] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0032] Figure 1 A schematic diagram showing the blurring of the entire image range provided by an embodiment of the present application;

[0033] Figure 2 A schematic diagram showing local shielding provided by an embodiment of the present application is shown;

[0034] Figure 3 A schematic diagram showing local glare provided by an embodiment of the present application is shown;

[0035] Figure 4 A schematic diagram showing the simultaneous existence of blur and glare provided by an embodiment of the present application is shown;

[0036] Figure 5 A schematic diagram of a flow chart of an image quality detection method provided by an embodiment of the present application is shown;

[0037] Figure 6 A schematic diagram of an image anomaly detection network provided by an embodiment of the present application is shown;

[0038] Figure 7 A schematic diagram of a feature recognition module provided in an embodiment of the present application is shown;

[0039] Figure 8 A schematic diagram of a feature extraction network provided in an embodiment of the present application is shown;

[0040] Fig. 9 A schematic diagram of a global anomaly detection network provided by an embodiment of the present application is shown;

[0041] Fig.10 A schematic diagram of a flow chart of an image quality detection method provided by an embodiment of the present application is shown;

[0042] Fig.11 A schematic diagram of an image quality detection device provided by an embodiment of the present application is shown;

[0043] Fig.12 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0044] Fig.13 A schematic diagram of a storage medium provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0045] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0046] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by technicians in the field to which this application belongs.

[0047] In the scenario of autonomous driving, there are many reasons for camera imaging problems. Common image quality problems in the scenario of autonomous driving are divided into three situations: occlusion, glare, and blur. Among them, occlusion refers to the complete invisibility of objects behind it caused by large water droplets, leaves, dirt, wipers, etc. Glare refers to the severe distortion of local areas caused by strong light sources such as car lights, street lights, and sunlight. Blur refers to the semi-transparent situation that is not completely invisible due to rain, fog, dust, etc. Usually, according to the size of the problem area distributed in the entire image, image quality problems can be divided into full-image quality problems and local quality problems. According to experience, occlusion and glare usually appear locally, while blur often covers the entire image. Therefore, blur is classified as a full-image quality problem, while occlusion and glare are classified as local quality problems.

[0048] Please refer to Figure 1-Figure 4 ,in Figure 1 It is a blur of the entire image. Figure 2 and Figure 3 They are local occlusion and glare respectively. It should be understood that an image may have both global blur and local occlusion and glare. Figure 4 is an example of blur and glare present simultaneously.

[0049] Image quality affects perception and decision-making processes, so it is particularly important to accurately identify image quality issues.

[0050] Traditional image quality diagnosis comprehensively evaluates the quality of images through manually defined evaluation indicators (such as signal-to-noise ratio, white balance, color difference and distortion, etc.). This method relies too much on manually defined evaluation indicators and is highly subjective. In addition, these manually defined evaluation indicators are usually not suitable for the application scenario of autonomous driving.

[0051] In the embodiment of the present application, an image acquisition device is provided on the device as the execution subject, and the image acquisition device includes but is not limited to a camera, a laser radar, etc. During the movement, the device acquires images of the surrounding environment through the image acquisition device, and uses the acquired images as images to be detected.

[0052] The sports scenes to which the embodiments of the present application can be applied include, but are not limited to, land traffic scenes, air traffic scenes, and water traffic scenes. In land traffic scenes, the devices as the execution subjects may include, but are not limited to, cars, motorcycles, electric vehicles, bicycles, pedestrians, etc. In air traffic scenes, the devices as the execution subjects may include, but are not limited to, various aircraft such as airplanes, helicopters, and drones. In water traffic scenes, the devices as the execution subjects may include, but are not limited to, various ships, jet skis, water bikes, etc.

[0053] In this embodiment, the motion mode of the device as the execution subject may be an autonomous driving mode or a semi-autonomous driving mode, etc. Among them, in the autonomous driving mode, the device may use the method provided in the embodiment of the present application to perform quality detection on the images collected by the image acquisition device. The semi-autonomous driving mode may be a driving mode in which both autonomous driving and human operation are involved. In the semi-autonomous driving mode, the device may also use the method provided in the embodiment of the present application to perform quality detection on the images collected by the image acquisition device.

[0054] like Figure 5 As shown, the method is applied to an image anomaly detection network, the image anomaly detection network includes a feature extraction network, a global anomaly detection network and a local anomaly detection network, the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; the method may include the following steps:

[0055] Step 501: extracting image features of the image to be detected through a feature extraction network;

[0056] Step 502: using a global anomaly detection network to identify image features, and obtaining parameters representing the degree of blur of the global blur of the image to be detected; and using a local anomaly detection network, and based on the image features, obtaining regional parameters representing the local quality abnormality area in the image to be detected, and marking the local quality abnormality area on the image to be detected according to the regional parameters.

[0057] Please refer to Figure 6 , Figure 6 The present invention is a schematic diagram of an image anomaly detection network applicable to the image quality detection method, wherein the image anomaly detection network includes a feature extraction network, a local anomaly detection network and a global anomaly detection network, wherein the output of the feature extraction network is respectively connected to the input of the local anomaly detection network and the input of the global fuzzy regression prediction network, the feature extraction network is used to extract image features of the input image, the local anomaly detection network obtains regional parameters representing local quality anomaly regions in the image to be detected based on the image features, and marks local quality anomaly regions on the image to be detected according to the regional parameters, and the global anomaly detection network identifies image features to obtain parameters representing the degree of blur of the global blur of the image to be detected.

[0058] In this embodiment, in order to make the network have a faster reasoning speed, a lightweight feature extraction network can be used to extract image features of the image to be detected. In the application, the feature extraction network includes but is not limited to MixVarGENet.

[0059] In some embodiments, in order to make the local anomaly detection network have better accuracy, the image features processed by the local anomaly detection network include semantic information features and texture information features.

[0060] Accordingly, the feature extraction network includes a semantic information extraction module and a texture information extraction module, and the image features include semantic information features extracted by the semantic information extraction module and texture information features extracted by the texture information extraction module; the local anomaly detection network includes a fusion module and a feature recognition module;

[0061] The local anomaly detection network is used, and based on the image features, the regional parameters characterizing the local quality anomaly area in the image to be detected are obtained, including:

[0062] The semantic information features and texture information features are fused through a fusion module to obtain fusion features;

[0063] The feature recognition module is used to identify the fused features and obtain the regional parameters.

[0064] In order to improve the accuracy of the model, the feature recognition module is implemented by three networks connected in parallel, and the three networks connected in parallel are used to identify the fusion features respectively to obtain the regional parameters.

[0065] In the specific implementation, the feature recognition module is used to identify the fusion features and obtain the regional parameters, including:

[0066] The fused features are identified by a first network to obtain a position parameter representing the position of the local quality abnormality region in the image to be detected; the fused features are identified by a second network to obtain a type parameter representing the abnormality type of the local quality abnormality region; and the fused features are identified by a third network to obtain a size parameter representing the size of the local quality abnormality region;

[0067] Determine the position parameter, type parameter, and size parameter as area parameters.

[0068] The abnormality types in this embodiment include but are not limited to types such as occlusion or glare.

[0069] In this embodiment, in order to simplify the network complexity of the first network, the second network and the third network, the first network, the second network and the third network are all implemented by multiple convolutional layers connected in series. It should be understood that the parameters of the convolutional layers in different networks may not be exactly the same.

[0070] In this embodiment, in order to reduce the resource consumption of the local anomaly detection network, when the first network, the second network and the third network all include multiple convolutional layers, the first network, the second network and the third network can share the convolutional layers with the same input and output.

[0071] See also Figure 7 , Figure 7 A schematic diagram of a feature recognition module shown in this embodiment. The figure includes 8 convolutional layers, namely convolutional layer 1-convolutional layer 8, wherein the first network is convolutional layer 1-convolutional layer 4 connected in series, the second network is convolutional layer 1, convolutional layer 2, convolutional layer 5 and convolutional layer 6 connected in series, and the third network is convolutional layer 1, convolutional layer 2, convolutional layer 7 and convolutional layer 8 connected in series.

[0072] In this embodiment, the semantic information features and texture information features are fused by data splicing, but not limited to. In specific implementation, the semantic information features and texture information features are fused by a fusion module to obtain fusion features, including:

[0073] Call the data connection function to concatenate the semantic information features and texture information features to obtain the fused features.

[0074] In an optional embodiment, the feature extraction network includes a convolution layer and a plurality of residual convolution layers connected to the convolution layer, and the plurality of residual convolution layers are connected in series. In order to improve the accuracy of the local anomaly detection network, the semantic information extraction module includes a first residual convolution layer, and the texture information extraction module includes a last residual convolution layer, and the first residual convolution layer is the first to be connected to the convolution layer among the plurality of residual convolution layers, and the last residual convolution layer is the last to be connected to the convolution layer among the plurality of residual convolution layers.

[0075] Please refer to Figure 8 , Figure 8 A schematic diagram of a feature extraction network shown in this embodiment. The feature extraction network includes a convolution layer and 5 residual convolution layers (res block), the convolution layer is the first network layer of the feature extraction network, the convolution layer and the 5 residual convolution layers are connected in series in sequence, the image to be detected is input into the feature extraction network from the convolution layer, and the image features extracted by the feature extraction network are output from the 5th residual convolution layer of the feature extraction network.

[0076] In an optional embodiment, based on Figure 8 The feature extraction network gives the following extraction process:

[0077] Input the image to be detected into the feature extraction network;

[0078] The first network layer of the feature extraction network is used to downsample the image to be detected to obtain the first network layer features of the image to be detected;

[0079] A convolution operation includes: using the i-th network layer of the feature extraction network to convolve the i-1-th network layer features of the image to be detected, to obtain the i-th network layer features of the image to be detected, where the value of i is greater than 1 and less than or equal to N, and N is the number of network layers of the feature extraction network;

[0080] When the i-th network layer is not the last network layer in the feature extraction network, update i to be equal to i+1, and return to perform a convolution operation until the features of each network layer of the image to be detected are obtained.

[0081] Still Figure 8 For example, in order to make the local anomaly detection network have better accuracy, the semantic information features processed by the local anomaly detection network can be the features output by the first residual convolutional network, and the texture information features can be the features output by the fifth residual convolutional network.

[0082] In an optional embodiment, if Fig. 9 As shown, the global anomaly detection network includes a pooling layer, a fully connected layer, and an activation function layer;

[0083] The pooling layer is connected to the feature extraction network and the fully connected layer respectively;

[0084] The fully connected layer is also connected to the activation function layer.

[0085] In this embodiment, a global anomaly detection network is used to identify image features to obtain parameters of the blur degree of the global blur of the image to be detected. It should be understood that this solution not only considers the global anomaly type through the global anomaly detection network, but also locates the local anomaly type through the local anomaly detection network, which can provide more sufficient information for downstream decision-making.

[0086] The technical solution provided in this embodiment extracts the image features of the image to be detected through a feature extraction network, and uses a global anomaly detection network and a local anomaly detection network to identify the image features respectively, thereby obtaining parameters characterizing the degree of blur of the global blur of the image to be detected and the local quality abnormality area of ​​the image to be detected. Compared with the traditional image quality diagnosis that relies on the selection of manual evaluation indicators, this application uses a deep neural network to directly evaluate the quality of the image, avoiding the subjectivity of manual features and improving the accuracy of quality detection.

[0087] The following is a specific example of the image quality detection method in the present application. Fig.10 , Fig.10A schematic diagram of an image anomaly detection network applicable to the image quality detection method is shown in FIG. Fig.10 In the figure, the feature extraction network includes a convolution layer and 5 residual convolution layers, and the local anomaly detection network includes 8 convolution layers and fusion modules, namely convolution layer 1-convolution layer 8. The first network is a series connection of convolution layer 1-convolution layer 4, the second network is a series connection of convolution layer 1, convolution layer 2, convolution layer 5 and convolution layer 6, and the third network is a series connection of convolution layer 1, convolution layer 2, convolution layer 7 and convolution layer 8. The global anomaly network includes a pooling layer, a fully connected layer and an activation function layer, wherein the global anomaly network obtains a regression value between 0 and 1, representing the degree of abnormality of the global blur of the image.

[0088] The embodiment of the present application also provides an image quality detection device, which is applied to an image anomaly detection network, wherein the image anomaly detection network includes a feature extraction network, a global anomaly detection network and a local anomaly detection network, wherein the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; Fig.11 As shown, the device comprises:

[0089] An extraction module 1101 is used to extract image features of the image to be detected through the feature extraction network;

[0090] The processing module 1102 is used to use the global anomaly detection network to identify the image features and obtain parameters that characterize the degree of blur of the global blur of the image to be detected; and use a local anomaly detection network and, based on the image features, obtain regional parameters that characterize the local quality abnormality area in the image to be detected, and mark the local quality abnormality area on the image to be detected according to the regional parameters.

[0091] In some embodiments, the feature extraction network includes a semantic information extraction module and a texture information extraction module, the image features include semantic information features extracted by the semantic information extraction module and texture information features extracted by the texture information extraction module; the local anomaly detection network includes a fusion module and a feature recognition module;

[0092] The processing module 1102 is used for:

[0093] The semantic information feature and the texture information feature are fused by the fusion module to obtain a fusion feature;

[0094] The feature recognition module is used to recognize the fusion feature to obtain the region parameter.

[0095] In some embodiments, the feature recognition module includes a first network, a second network, and a third network connected in parallel; the processing module 1102 is used to:

[0096] The first network is used to identify the fused features to obtain a position parameter representing the position of the local quality abnormality region in the image to be detected; the second network is used to identify the fused features to obtain a type parameter representing the abnormality type of the local quality abnormality region; and the third network is used to identify the fused features to obtain a size parameter representing the size of the local quality abnormality region;

[0097] The position parameter, the type parameter, and the size parameter are determined as the region parameters.

[0098] In some embodiments, the first network, the second network, and the third network each include a plurality of convolutional layers connected in series.

[0099] In some embodiments, the feature extraction network includes a convolutional layer and multiple residual convolutional layers connected to the convolutional layer, and the multiple residual convolutional layers are connected in series; the semantic information extraction module includes a first residual convolutional layer, and the texture information extraction module includes a last residual convolutional layer, the first residual convolutional layer is the first to be connected to the convolutional layer among the multiple residual convolutional layers, and the last residual convolutional layer is the last to be connected to the convolutional layer among the multiple residual convolutional layers.

[0100] In some embodiments, the fusion module includes a preset data connection function;

[0101] The processing module 1102 is used for:

[0102] The data connection function is called to concatenate the semantic information feature and the texture information feature to obtain the fusion feature.

[0103] In some embodiments, the global anomaly detection network includes a pooling layer, a fully connected layer, and an activation function layer;

[0104] The pooling layer is connected to the feature extraction network and the fully connected layer respectively;

[0105] The fully connected layer is also connected to the activation function layer.

[0106] The image quality detection device provided in the embodiment of the present application and the image quality detection method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0107] The present application also provides an electronic device to perform the above-mentioned image quality detection method. Fig.12 It shows a schematic diagram of an electronic device provided by some embodiments of the present application. Fig.12As shown, the electronic device 12 includes: a processor 1200, a memory 1201, a bus 1202 and a communication interface 1203, and the processor 1200, the communication interface 1203 and the memory 1201 are connected via the bus 1202; the memory 1201 stores a computer program that can be run on the processor 1200, and the processor 1200 executes the image quality detection method provided in any of the aforementioned embodiments of the present application when running the computer program.

[0108] The memory 1201 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the device network element and at least one other network element is realized through at least one communication interface 1203 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0109] The bus 1202 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 1201 is used to store programs, and the processor 1200 executes the program after receiving an execution instruction. The image quality detection method disclosed in any implementation of the embodiment of the present application may be applied to the processor 1200, or implemented by the processor 1200.

[0110] The processor 1200 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 1200. The above processor 1200 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 1201, and the processor 1200 reads the information in the memory 1201 and completes the steps of the above method in combination with its hardware.

[0111] The electronic device provided in the embodiment of the present application and the image quality detection method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0112] The present application also provides a computer-readable storage medium corresponding to the image quality detection method provided in the above embodiment. Fig.13 The computer-readable storage medium shown is a CD 30 on which a computer program (ie, a program product) is stored. When the computer program is run by a processor, the image quality detection method provided in any of the aforementioned embodiments will be executed.

[0113] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0114] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the image quality detection method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0115] It should be noted that:

[0116] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0117] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following schematic diagram: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0118] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0119] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An image quality detection method, characterized in that: The method is applied to an image anomaly detection network, the image anomaly detection network includes a feature extraction network, a global anomaly detection network and a local anomaly detection network, the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; the method includes: Extracting image features of the image to be detected through the feature extraction network; The global anomaly detection network is used to identify the image features to obtain parameters characterizing the degree of blur of the global blur of the image to be detected; and a local anomaly detection network is used to obtain regional parameters characterizing the local quality abnormality area in the image to be detected based on the image features, and the local quality abnormality area is marked on the image to be detected according to the regional parameters.

2. The method according to claim 1, characterized in that The feature extraction network includes a semantic information extraction module and a texture information extraction module, the image features include semantic information features extracted by the semantic information extraction module and texture information features extracted by the texture information extraction module; the local anomaly detection network includes a fusion module and a feature recognition module; Using a local anomaly detection network and based on the image features, obtaining regional parameters characterizing the local quality anomaly region in the image to be detected, including: The semantic information feature and the texture information feature are fused by the fusion module to obtain a fusion feature; The feature recognition module is used to recognize the fusion feature to obtain the region parameter.

3. The method according to claim 2, characterized in that The feature recognition module includes a first network, a second network, and a third network connected in parallel; the feature recognition module is used to recognize the fusion feature to obtain the region parameter, including: The first network is used to identify the fused features to obtain a position parameter representing the position of the local quality abnormality region in the image to be detected; the second network is used to identify the fused features to obtain a type parameter representing the abnormality type of the local quality abnormality region; and the third network is used to identify the fused features to obtain a size parameter representing the size of the local quality abnormality region; The position parameter, the type parameter, and the size parameter are determined as the region parameters.

4. The method according to claim 3, characterized in that The first network, the second network and the third network each include a plurality of convolutional layers connected in series.

5. The method according to any one of claims 2 to 4, characterized in that: The feature extraction network includes a convolution layer and multiple residual convolution layers connected to the convolution layer, and the multiple residual convolution layers are connected in series; the semantic information extraction module includes a first residual convolution layer, and the texture information extraction module includes a last residual convolution layer, the first residual convolution layer is the first to be connected to the convolution layer among the multiple residual convolution layers, and the last residual convolution layer is the last to be connected to the convolution layer among the multiple residual convolution layers.

6. The method according to any one of claims 2 to 4, characterized in that: The fusion module includes a preset data connection function; The fusion module is used to fuse the semantic information feature and the texture information feature to obtain a fusion feature, including: The data connection function is called to concatenate the semantic information feature and the texture information feature to obtain the fusion feature.

7. The method according to claim 1, characterized in that The global anomaly detection network includes a pooling layer, a fully connected layer and an activation function layer; The pooling layer is connected to the feature extraction network and the fully connected layer respectively; The fully connected layer is also connected to the activation function layer.

8. An image quality detection device, characterized in that: Applied to an image anomaly detection network, the image anomaly detection network includes a feature extraction network, a global anomaly detection network and a local anomaly detection network, the global anomaly detection network and the local anomaly detection network are respectively connected to the feature extraction network; the device includes: An extraction module, used for extracting image features of the image to be detected through the feature extraction network; A processing module is used to use the global anomaly detection network to identify the image features and obtain parameters that characterize the degree of blur of the global blur of the image to be detected; and, use a local anomaly detection network and, based on the image features, obtain regional parameters that characterize the local quality abnormality area in the image to be detected, and mark the local quality abnormality area on the image to be detected according to the regional parameters.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor runs the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to any one of claims 1 to 7.

Images