CN111340140A - Image data set acquisition method and device, electronic equipment and storage medium - Google Patents
Image data set acquisition method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111340140A CN111340140A CN202010239580.5A CN202010239580A CN111340140A CN 111340140 A CN111340140 A CN 111340140A CN 202010239580 A CN202010239580 A CN 202010239580A CN 111340140 A CN111340140 A CN 111340140A
- Authority
- CN
- China
- Prior art keywords
- image
- target object
- target
- object area
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention provides an image data set acquisition method, an image data set acquisition device, electronic equipment and a storage medium, wherein the method comprises the following steps: extracting a target object area image containing only a target object from each original image in the original image data set; inputting the target object area images into a quality evaluation model to obtain target quality values; and taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image to form a target image data set containing the target object. Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, and the obtained target image data set has higher quality.
Description
Technical Field
The present invention relates to the field of image classification technologies, and in particular, to a method and an apparatus for acquiring an image dataset, an electronic device, and a storage medium.
Background
Currently, with the development of artificial intelligence, image classification has been widely used as an important field of artificial intelligence. In image classification in the related art, an image data set of each target object may be obtained for each target object, a neural network model is trained using the image data set of each target object, and the trained neural network model is used as a classification model. And inputting the image to be classified into a classification model to obtain a classification result output by the classification model. In this way, image classification is achieved.
In the image classification process of the related art, if the classification model itself is not accurate enough, the classification effect may be biased. The reason why the classification model itself is not accurate enough is mainly as follows: the image quality of the images in the image dataset from which the neural network model is trained is not high enough. Therefore, it is desirable to ensure the image quality of the images in the image dataset in order for the classification model to be more accurate.
At present, in the related art, in order to ensure the image quality of an image in an image data set, a manner of obtaining the image data set including a target object for the target object is generally to acquire a large number of original images including the target object, select an original image with a larger resolution, that is, a clearer original image, according to the size of the resolution, and add the selected original image into the image data set of the target object.
For example: if the target object is an automobile, a large number of original images containing the automobile are collected, then the clear original images are selected according to the resolution and added into an image data set of the automobile.
However, the inventor of the present application finds in practice that in such a manner that the original image is added to the target image data set to obtain the target image data set only according to the size of the resolution, the image quality of the finally obtained image in the target image data set still needs to be improved.
Disclosure of Invention
An object of the embodiments of the present invention is to provide an image dataset acquiring method, an image dataset acquiring device, an electronic device, and a storage medium, so as to improve image quality of an image in a target image dataset. The specific technical scheme is as follows:
in a first aspect, an embodiment of the present invention provides a method for acquiring an image data set, where the method includes:
obtaining an original image dataset containing a target object; the raw image dataset comprises a plurality of raw images;
extracting a target object area image containing only a target object from each original image in the original image data set;
inputting each target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
and taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image to form a target image data set containing the target object.
Optionally, before the step of inputting each target object area image into the quality evaluation model and obtaining the target quality value of each target object area image output by the quality evaluation model, the method further includes:
carrying out image transformation processing on each target object area image to obtain a corresponding target transformation image; the image transformation processing is used for increasing the number of target object area images and/or increasing the resolution;
the step of inputting the target object area images into a quality evaluation model and acquiring target quality values of the target object area images output by the quality evaluation model comprises the following steps:
and inputting the obtained target transformation images and the target object area images into a quality evaluation model as target object area images, and obtaining target quality values of the target object area images output by the quality evaluation model.
Optionally, the step of extracting a target object region image only including a target object from each original image in the original image data set includes:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as target object area images; the image recognition network model is as follows: and training a preset second deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance.
Optionally, the step of extracting a target object region image only including a target object from each original image in the original image data set includes:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as images to be determined; the image recognition network model is as follows: the method comprises the steps that an original sample image containing a target object and an image of a region where the target object is located in the original sample image are used as marked sample images in advance, and a preset third deep learning network model is trained to obtain the marked sample images;
and acquiring an image with the resolution being greater than a preset resolution threshold and/or the noise intensity value being less than a preset noise intensity threshold in the image to be determined as a target object area image.
Optionally, after the step of acquiring, as the target object region image, the image in which the resolution in the image to be determined is greater than the preset resolution threshold and/or the noise intensity value is less than the preset noise intensity threshold, the method further includes:
acquiring the definition of the residual images of the target object region removed from the plurality of images to be determined;
taking the residual image with the definition greater than a preset definition threshold value as a supplementary target object area image;
the step of inputting the target object area images into a quality evaluation model and acquiring target quality values of the target object area images output by the quality evaluation model comprises the following steps:
and inputting each target object area image and each supplemented target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model.
Optionally, the step of performing image transformation processing on each target object region image to obtain a corresponding target transformation image includes:
performing multiple data enhancement processing on each target object region image by using a preset image data enhancement algorithm to obtain multiple target transformation images; and/or the presence of a gas in the gas,
based on an image transformation network model, enhancing or suppressing different characteristics in the target object area image, and obtaining a target transformation image corresponding to the target object area image; and/or the presence of a gas in the gas,
and inputting each target object area image into a preset super-resolution network model, and obtaining a target transformation image which is output by the preset super-resolution network model and has higher resolution than the target object area image.
Optionally, the training process of the quality evaluation model includes:
acquiring a training sample set; the training sample set comprises a plurality of training samples; wherein each training sample comprises: sample images and image quality values marked by the sample images;
inputting each sample image in the training sample set into a current first deep learning network model, and acquiring a sample quality value corresponding to each sample image output by the current first deep learning network model;
calculating a loss value based on each sample intrinsic quantity value, each labeled image quality value and a preset loss function;
judging whether the current first deep learning network model converges or not according to a loss value of a preset loss function; if so, taking the current first deep learning network model as a trained quality evaluation model; and if not, adjusting and updating the network parameters of the current first deep learning network model, and returning to the step of inputting each sample image in the training sample set into the current first deep learning network model and acquiring the sample quality value corresponding to each sample image output by the current first deep learning network model.
In a second aspect, an embodiment of the present invention provides an apparatus for acquisition of an image dataset, the apparatus including:
an original image data set obtaining unit for obtaining an original image data set containing a target object; the raw image dataset comprises a plurality of raw images;
a target object region image extracting unit configured to extract a target object region image containing only a target object from each original image in the original image data set;
a target quality value acquisition unit, configured to input each target object area image to a quality evaluation model, and acquire a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
and the target image data set forming unit is used for forming a target image data set containing a target object by taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image.
Optionally, the apparatus further comprises:
a target transformation image obtaining unit, configured to perform image transformation processing on each target object region image to obtain a corresponding target transformation image before the target quality value obtaining unit inputs each target object region image into the quality evaluation model and obtains a target quality value of each target object region image output by the quality evaluation model; the image transformation processing is used for increasing the number of target object area images and/or increasing the resolution;
the target quality value obtaining unit is specifically configured to:
and inputting the obtained target transformation images and the target object area images into a quality evaluation model as target object area images, and obtaining target quality values of the target object area images output by the quality evaluation model.
Optionally, the target object region image extracting unit is specifically configured to:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as target object area images; the image recognition network model is as follows: and training a preset second deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance.
Optionally, the target object region image extracting unit includes:
the to-be-determined image acquisition module is used for respectively inputting each original image in the original image data set to a pre-trained image recognition network model and acquiring a plurality of images output by the image recognition network model as to-be-determined images; the image recognition network model is as follows: the method comprises the steps that an original sample image containing a target object and an image of a region where the target object is located in the original sample image are used as marked sample images in advance, and a preset third deep learning network model is trained to obtain the marked sample images;
and the target object area image acquisition module is used for acquiring an image of which the resolution is greater than a preset resolution threshold and/or the noise intensity value is less than a preset noise intensity threshold in the image to be determined as a target object area image.
Optionally, the apparatus further comprises:
a residual image sharpness obtaining unit, configured to obtain, after the target object region image obtaining module obtains, as a target object region image, an image in the to-be-determined image with a resolution greater than a preset resolution threshold and/or a noise intensity value smaller than a preset noise intensity threshold, sharpness of a residual image of the target object region image being removed from the plurality of to-be-determined images;
a supplementary target object area image acquisition unit, configured to take a remaining image with a definition greater than a preset definition threshold as a supplementary target object area image;
the target object area image acquisition module is specifically configured to:
and inputting each target object area image and each supplemented target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model.
Optionally, the target transform image obtaining unit is specifically configured to:
performing multiple data enhancement processing on each target object region image by using a preset image data enhancement algorithm to obtain multiple target transformation images; and/or the presence of a gas in the gas,
based on an image transformation network model, enhancing or suppressing different characteristics in the target object area image, and obtaining a target transformation image corresponding to the target object area image; and/or the presence of a gas in the gas,
and inputting each target object area image into a preset super-resolution network model, and obtaining a target transformation image which is output by the preset super-resolution network model and has higher resolution than the target object area image.
Optionally, the apparatus further includes: a quality evaluation model training unit;
the quality evaluation model training unit is specifically configured to:
acquiring a training sample set; the training sample set comprises a plurality of training samples; wherein each training sample comprises: sample images and image quality values marked by the sample images;
inputting each sample image in the training sample set into a current first deep learning network model, and acquiring a sample quality value corresponding to each sample image output by the current first deep learning network model;
calculating a loss value based on each sample intrinsic quantity value, each labeled image quality value and a preset loss function;
judging whether the current first deep learning network model converges or not according to a loss value of a preset loss function; if so, taking the current first deep learning network model as a trained quality evaluation model; and if not, adjusting and updating the network parameters of the current first deep learning network model, and returning to the step of inputting each sample image in the training sample set into the current first deep learning network model and acquiring the sample quality value corresponding to each sample image output by the current first deep learning network model.
In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor and the communication interface complete communication between the memory and the processor through the communication bus;
a memory for storing a computer program;
and a processor for implementing the method steps for acquiring any of the image data sets described above when executing the program stored in the memory.
In a fourth aspect, the embodiment of the present invention provides a computer-readable storage medium, in which a computer program is stored, where the computer program is executed by a processor to perform the steps of any one of the above-mentioned image data set acquisition methods.
In a fifth aspect, embodiments of the present invention further provide a computer program product containing instructions, which when run on a computer, cause the computer to execute any one of the above-mentioned image data set acquisition methods.
According to the method, the device, the electronic equipment and the storage medium for acquiring the image data set, which are provided by the embodiment of the invention, the original image data set containing the target object can be acquired; the raw image dataset comprises a plurality of raw images; extracting a target object area image containing only a target object from each original image in the original image data set; inputting each target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image; and taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image to form a target image data set containing the target object.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
Of course, not all of the advantages described above need to be achieved at the same time in the practice of any one product or method of the invention.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a method of image dataset acquisition provided by an embodiment of the present invention;
FIG. 2a is an exemplary diagram of an original image;
FIG. 2b is an exemplary diagram of a target object area image extracted from the exemplary diagram of the original image shown in FIG. 2 a;
FIG. 3 is another flow chart of a method of acquisition of an image dataset provided by an embodiment of the invention;
FIG. 4 is a flow chart of yet another method of image dataset acquisition provided by an embodiment of the present invention;
FIG. 5 is a detailed flowchart of a second image transformation method of step S303 in the embodiment shown in FIG. 3 and step S406 in the embodiment shown in FIG. 4;
FIG. 6 is a flow chart of a method of training a quality assessment model used in embodiments of the present invention;
FIG. 7 is a schematic structural diagram of an apparatus for acquiring an image dataset according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to improve the image quality of an image in a target image dataset, the embodiment of the invention provides an image dataset acquisition method, an image dataset acquisition device, an electronic device and a storage medium.
The method for acquiring the image data set provided by the embodiment of the invention can be applied to any electronic equipment which needs to improve the image quality of the image in the target image data set, such as: a computer or a mobile terminal, etc., which are not limited herein. For convenience of description, the electronic device is hereinafter referred to simply as an electronic device.
Referring to fig. 1, for a method for acquiring an image data set according to an embodiment of the present invention, as shown in fig. 1, a specific processing flow of the method may include:
step S101, obtaining an original image data set containing a target object; the raw image dataset contains a plurality of raw images.
Step S102, extracting a target object area image containing only a target object from each original image in the original image data set.
Step S103, inputting each target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model.
Wherein the quality evaluation model is as follows: and training a preset first deep learning network model by using the sample image and the marked image quality value thereof.
And step S104, taking the target object area image with the target quality value exceeding the preset quality value threshold as a target image, and forming a target image data set containing the target object.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
It can be implemented that the examples of the original image and the target object area image described in the above embodiments can be as shown in fig. 2a and fig. 2b, respectively.
Referring to fig. 2a, an exemplary diagram of an original image, including a target object and a background portion;
referring to fig. 2b, an exemplary diagram of the target object area image extracted from the exemplary diagram of the original image shown in fig. 2a is shown in fig. 2b, that is, the target object area image is: and a rectangular frame image formed by the area where the target object is located.
Further, the step S102 has at least the following two embodiments:
the first embodiment: respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as target object area images; the image recognition network model is as follows: and training a preset second deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance.
The specific training process of the image recognition network model can be implemented as follows: acquiring an original sample image containing a target object and an image of an area where the target object is located in the original sample image as an annotated sample image; inputting the original sample image into a current second deep learning network model, and acquiring an image output by the current second deep learning network model as a target object region predicted image; calculating a loss value based on each target object region prediction image, each annotation sample image and a preset loss function; judging whether the current second deep learning network model is converged or not according to a loss value of a preset loss function; if so, taking the current second deep learning network model as a trained image recognition network model; and if not, adjusting and updating the network parameters of the current second deep learning network model, returning to execute the step of inputting the original sample image into the current second deep learning network model, and acquiring the image output by the current second deep learning network model as a target object region predicted image.
For the second deep learning network model, the preset loss function may be a square loss function, and the specific formula is as follows:
wherein Loss2 is a Loss value of a preset Loss function; n is the serial number of the original sample image; s is the number of original sample images; f'nLabeling a sample image corresponding to the nth original sample image; fnAnd predicting the image for the target object area corresponding to the nth original sample image.
It can be implemented that, for the second deep learning network model, the preset loss function can also be other loss functions, for example: the minimum absolute deviation L1 loss function is not specifically limited herein.
The second embodiment: respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as images to be determined; the image recognition network model is as follows: the method comprises the steps that an original sample image containing a target object and an image of a region where the target object is located in the original sample image are used as marked sample images in advance, and a preset third deep learning network model is trained to obtain the marked sample images; and acquiring an image with the resolution being greater than a preset resolution threshold and/or the noise intensity value being less than a preset noise intensity threshold in the image to be determined as a target object area image.
The third deep learning network model can be the same as the second deep learning network model in training method.
By adopting the second embodiment, the plurality of images output by the image recognition network model are screened, and the images with the resolution greater than the preset resolution threshold and/or the noise intensity value less than the preset noise intensity threshold are screened out and serve as the target object area images. Thus, the image quality of the obtained target object region image is high.
Referring to fig. 3, another flowchart of a method for acquiring an image dataset according to an embodiment of the present invention includes:
step S301, obtaining an original image data set containing a target object; the raw image dataset contains a plurality of raw images.
Step S302, extracting a target object region image containing only a target object from each original image in the original image data set.
Step S303, carrying out image transformation processing on each target object region image to obtain a corresponding target transformation image; the image transformation process is used to increase the number of target object region images and/or increase the resolution.
Step S304, all the obtained target transformation images and all the obtained target object area images are used as target object area images and input into a quality evaluation model, and target quality values of all the target object area images output by the quality evaluation model are obtained.
Wherein the quality evaluation model is as follows: and training a preset first deep learning network model by using the sample image and the marked image quality value thereof.
Step S305, using the target object region image whose target quality value exceeds the preset quality value threshold as a target image, and forming a target image data set including the target object.
Therefore, the target transformation image and each target object area image are used as the target object area images, the number of the target object area images is increased, the efficiency of obtaining the target object area images is improved, the efficiency of obtaining the target images is further improved, and the time for constructing the target image data set is reduced.
Referring to fig. 4, a flowchart of a method for acquiring an image data set according to an embodiment of the present invention includes:
step S401, obtaining an original image data set containing a target object; the raw image dataset contains a plurality of raw images.
It may be implemented that the original image in the original image data set may be a video frame image or an image that is obtained by a crawler and contains the target object, or a video frame image or an image that is collected by a high definition camera and the like and contains the target object, and herein, the obtaining manner of the original image data set is not particularly limited.
Step S402, inputting each original image in the original image data set to a pre-trained image recognition network model respectively, and acquiring a plurality of images output by the image recognition network model as images to be determined.
In practice, the image recognition network model is: and training a preset third deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance. Specifically, the preset third deep learning network model is a preset convolutional neural network model.
Step S403, acquiring an image of the image to be determined, in which the resolution is greater than a preset resolution threshold and/or the noise intensity value is less than a preset noise intensity threshold, as a target object region image.
The method specifically includes the following three implementation manners.
In a first mode, the electronic device may acquire an image with a resolution greater than a preset resolution threshold in the image to be determined, as a target object region image.
In practical application, the image with the resolution less than or equal to the preset resolution threshold has the problem of too low definition, and in this way, the image with the lower resolution is filtered, the image quality of the obtained target object region image is higher, and more target images with higher quality can be obtained.
In a second mode, the electronic device may obtain an image, of which the noise intensity value is smaller than a preset noise intensity threshold value, in the image to be determined as a target object area image.
In practical application, the image with the noise characteristic value being greater than or equal to the preset intensity threshold has the problems of overlarge noise or over-blurring and the like, and by the method, the image with the larger noise intensity value is removed, the image quality of the obtained target object area image is higher, and then more target images with higher quality can be obtained.
In a third mode, the electronic device may obtain, as the target object region image, an image in which the resolution is greater than a preset resolution threshold and the noise intensity value is less than a preset noise intensity threshold in the image to be determined.
By adopting the mode, the resolution ratio of the target object area image is greater than the preset resolution ratio threshold value, and meanwhile, the noise intensity value is required to be smaller than the preset noise intensity threshold value, so that the screening condition of the target object area image is stricter, and the quality of the obtained target object area image is higher.
Step S404, acquiring the definition of the residual images of the target object area removed from the plurality of images to be determined.
The definition represents the definition of an image, and the electronic equipment can judge whether the image is clear or not through the pixel value gradient or the contrast.
What may be implemented, for example: the target object is an automobile, and the target object area image is a rectangular frame image formed by areas where the automobile is located.
The electronic device can judge whether the area where the pixel points forming the automobile are located is clear or not by judging the pixel value gradient of the pixel points forming the automobile in the residual image. If the pixel value gradient of the pixel point at the junction of the area where the pixel points forming the automobile are located and the area where the pixel points not forming the automobile are located is greater than the preset gradient threshold value, the area where the pixel points forming the automobile are located is a clear area.
The electronic device may also determine that the area where the pixel points constituting the car are located is a clear area when the contrast of the area where the pixel points constituting the car are located is greater than a preset contrast threshold.
And step S405, taking the residual image with the definition larger than a preset definition threshold value as a supplementary target object area image.
Corresponding to the above step S404, an image in which the area where the pixel points constituting the automobile are located is clear is taken as a supplementary target object area image.
Step S406, carrying out image transformation processing on each target object area image and the supplemented target object area image to obtain corresponding target transformation images; the image transformation process is used to increase the number of target object region images and/or increase the resolution.
It may be implemented that the image transformation aims at changing the image such that the changed image is distinguished from the image before the change.
In this embodiment, each target object region image and the supplemented target object region image are subjected to image transformation processing to obtain a corresponding target transformation image; in other embodiments, only the target object area image may be subjected to image transformation processing to obtain a corresponding target transformed image, or only the supplemented target object area image may be subjected to image transformation processing to obtain a corresponding target transformed image.
Step S407, inputting each obtained target transformation image, each target object region image, and the supplemented target object region image as target object region images to a quality evaluation model, and obtaining a target quality value of each target object region image output by the quality evaluation model.
In this embodiment, each of the obtained target transformation images, each of the obtained target object region images, and the obtained supplementary target object region image are input to the quality evaluation model as target object region images; in other embodiments, the target transformation image, the target object region image, and the supplemented target object region image may be input to the quality evaluation model as the target object region image, alone or in any combination.
Step S408, using the target object region image whose target quality value exceeds the preset quality value threshold as a target image, and forming a target image data set including the target object.
For example: the target mass value may range from 0 to 100 and the preset mass value threshold may be 85.
By adopting the mode, the target object area image with the target quality number value exceeding the preset quality number value threshold is used as the target image, and the obtained target image has higher quality.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
Meanwhile, each target object area image and/or the supplemented target object area image is subjected to image transformation processing to obtain a corresponding target transformation image, and the target transformation image is also used as the target object area image, so that the number of the target object area images is obviously greater than that of the original images, and the efficiency of constructing a target image data set is improved. By applying the embodiment of the invention, the target image with higher quality can be obtained more quickly.
In an embodiment of the present invention, the target object may be an automobile, so that the method of the embodiment of the present invention may obtain an automobile region image with a higher quality, where the automobile region image includes the automobile, and further train the preset neural network model using the automobile region image of the embodiment of the present invention, so that the obtained classification model is more accurate. Meanwhile, the corresponding target transformation image is obtained by the image transformation processing method and is also used as the automobile area image, so that the efficiency of obtaining the original image containing the automobile is saved, and the training speed of the classification model is accelerated.
Specifically, the specific implementation manner of performing image transformation processing on each target object region image to obtain a corresponding target transformation image in step S303 in the embodiment shown in fig. 3 and step S406 in the embodiment shown in fig. 4 at least includes the following three image transformation manners:
first image conversion method: and performing data enhancement processing on each target object region image for multiple times by using a preset image data enhancement algorithm to obtain multiple target transformation images.
The image data enhancement algorithm can be implemented as follows: more data which is as effective as the limited data is generated, and the distribution of training data is enriched. For example: and the data enhancement algorithms such as turning, cutting, rotating, shielding and the like can be adopted.
Second image conversion method: and enhancing or inhibiting different characteristics in the target object area image based on an image transformation network model to obtain a target transformation image corresponding to the target object area image.
The third image conversion method: and inputting each target object area image into a preset super-resolution network model, and obtaining a target transformation image which is output by the preset super-resolution network model and has higher resolution than the target object area image.
The preset super-resolution network model can be a super-resolution network model trained in the related art. A convolutional neural network model that converts a low-resolution image to a high-resolution image may be used. And then the target transformation image is also used as the target object area image. By adopting the image transformation mode, the number of the target object area images is increased, and simultaneously, the resolution ratio of the target object area images is increased, so that more target images with higher quality can be obtained more quickly.
In practice, each target object region image is subjected to image transformation processing to obtain a specific implementation of a corresponding target transformation image, and any combination of the above three image transformation modes can be adopted. For example: processing each target object area image in the three image transformation modes; processing each target object area image by only adopting any two of the three image transformation modes; or each target object area image is processed only by adopting any one of the image transformation modes.
As shown in fig. 5, a specific flow of the second image transformation method in step S303 in the embodiment shown in fig. 3 and step S406 in the embodiment shown in fig. 4 includes:
step S501, a pre-trained image transformation network model is obtained.
The image transformation network model is a generation network in a generation type countermeasure network trained by using a sample random vector and a target object area image, and the image transformation network model is a network model for outputting an image with high similarity to the target object area image after the random vector is input. For example: the image transformation network model may be a StyleGAN network model.
Step S502, a target random vector is obtained and input to the image transformation network model, and a target image output by the image transformation network model is obtained.
It may be implemented that the target random vector is generated by the electronic device.
Step S503, acquiring a region to be enhanced or suppressed in the target image as a target region by using an image segmentation method.
For example: the target object is an automobile, and the target object region image is an image including only the target object. The region to be enhanced is the headlight region in the image. Then the following steps can be adopted to enhance the car light region, and a target transformation image with the enhanced car light region is obtained; if the area to be suppressed is a background area other than a car, the following steps may be taken to suppress the background area.
Step S504, acquiring the image transformation network model, and using each intermediate feature map as each feature image when generating the target image.
And step S505, calculating the similarity of each feature image and the target area, and taking the feature image with the highest similarity as the target feature image.
Step S506, determining a target layer of the target characteristic image output in the image transformation network model, and adjusting parameters of an activation function in the target layer to enhance or suppress a target area to obtain a target transformation image.
The parameters of the activation function in the target layer are adjusted specifically by a received adjustment instruction or based on a preset adjustment rule.
The first mode is as follows: the method for adjusting the parameters through the received adjustment instruction may specifically be: in step S504, after each feature image is acquired, the feature image is subjected to upsampling visualization processing, and is displayed to the user, and the user inputs an adjustment instruction based on the displayed feature image.
The second mode is as follows: the parameter adjustment is carried out based on a preset adjustment rule, wherein the preset adjustment rule can be a preset parameter adjusting function, the electronic equipment can adjust the parameter of the activation function of the target layer based on the preset parameter adjusting function, manual participation is not needed, and automatic parameter adjustment is achieved.
The training process of the quality evaluation model in the above embodiment can be specifically referred to fig. 6.
As shown in fig. 6, which is a flowchart of a method for training a quality evaluation model used in an embodiment of the present invention, a specific processing flow of the method may include:
step S601, acquiring a training sample set; the training sample set comprises a plurality of training samples; wherein each training sample comprises: the sample image and the labeled image quality value thereof.
Step S602, inputting each sample image in the training sample set into the current first deep learning network model, and obtaining a sample quality value corresponding to each sample image output by the current first deep learning network model.
The current first deep learning network model may be a preset convolutional neural network model, and specifically may be a concept-50 network model.
Step S603, calculating a loss value based on each sample intrinsic quality value, each labeled image quality value, and a preset loss function.
In an implementation manner, for the first deep learning network model, the preset loss function may be an average absolute error loss function, and the specific formula is as follows:
wherein Loss1 is a Loss value of a preset Loss function; i is the serial number of the sample image in the training sample set; m is the number of sample images in the training sample set; y'iThe image quality value of the label corresponding to the ith sample image in the training sample set; y isiAnd the sample quality quantity value corresponding to the ith sample image in the training sample set.
It can be implemented that, for the first deep learning network model, the preset loss function can also be other loss functions, such as: the minimum absolute deviation L1 loss function is not specifically limited herein.
Step S604, determining whether the current first deep learning network model converges according to a loss value of a preset loss function.
If the judgment result is no, that is, the current first deep learning network model is not converged, executing step S605; if the result of the determination is yes, that is, the current first deep learning network model converges, step S606 is executed.
And step S605, adjusting and updating the network parameters of the current first deep learning network model. The process returns to step S602.
And step S606, taking the current first deep learning network model as a trained quality evaluation model.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
As shown in fig. 7, the schematic structural diagram of an apparatus for acquiring an image data set according to an embodiment of the present invention includes:
an original image data set obtaining unit 701 for obtaining an original image data set containing a target object; the raw image dataset comprises a plurality of raw images;
a target object region image extracting unit 702 configured to extract a target object region image containing only a target object from each original image in the original image data set;
a target quality number acquiring unit 703, configured to input each target object region image into a quality evaluation model, and acquire a target quality number of each target object region image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
a target image data set constructing unit 704 configured to construct a target image data set including the target object, with the target object region image whose target quality value exceeds a preset quality value threshold as the target image.
Optionally, the apparatus further comprises:
a target transformation image obtaining unit, configured to perform image transformation processing on each target object region image to obtain a corresponding target transformation image before the target quality value obtaining unit inputs each target object region image into the quality evaluation model and obtains a target quality value of each target object region image output by the quality evaluation model; the image transformation processing is used for increasing the number of target object area images and/or increasing the resolution;
the target quality value obtaining unit is specifically configured to:
and inputting the obtained target transformation images and the target object area images into a quality evaluation model as target object area images, and obtaining target quality values of the target object area images output by the quality evaluation model.
Optionally, the target object region image extracting unit is specifically configured to:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as target object area images; the image recognition network model is as follows: and training a preset second deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance.
Optionally, the target object region image extracting unit includes:
the to-be-determined image acquisition module is used for respectively inputting each original image in the original image data set to a pre-trained image recognition network model and acquiring a plurality of images output by the image recognition network model as to-be-determined images; the image recognition network model is as follows: the method comprises the steps that an original sample image containing a target object and an image of a region where the target object is located in the original sample image are used as marked sample images in advance, and a preset third deep learning network model is trained to obtain the marked sample images;
and the target object area image acquisition module is used for acquiring an image of which the resolution is greater than a preset resolution threshold and/or the noise intensity value is less than a preset noise intensity threshold in the image to be determined as a target object area image.
Optionally, the apparatus further comprises:
a residual image sharpness obtaining unit, configured to obtain, after the target object region image obtaining module obtains, as a target object region image, an image in the to-be-determined image with a resolution greater than a preset resolution threshold and/or a noise intensity value smaller than a preset noise intensity threshold, sharpness of a residual image of the target object region image being removed from the plurality of to-be-determined images;
a supplementary target object area image acquisition unit, configured to take a remaining image with a definition greater than a preset definition threshold as a supplementary target object area image;
the target object area image acquisition module is specifically configured to:
and inputting each target object area image and each supplemented target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model.
Optionally, the target transform image obtaining unit is specifically configured to:
performing multiple data enhancement processing on each target object region image by using a preset image data enhancement algorithm to obtain multiple target transformation images; and/or the presence of a gas in the gas,
based on an image transformation network model, enhancing or suppressing different characteristics in the target object area image, and obtaining a target transformation image corresponding to the target object area image; and/or the presence of a gas in the gas,
and inputting each target object area image into a preset super-resolution network model, and obtaining a target transformation image which is output by the preset super-resolution network model and has higher resolution than the target object area image.
Optionally, the apparatus further includes: a quality evaluation model training unit;
the quality evaluation model training unit is specifically configured to:
acquiring a training sample set; the training sample set comprises a plurality of training samples; wherein each training sample comprises: sample images and image quality values marked by the sample images;
inputting each sample image in the training sample set into a current first deep learning network model, and acquiring a sample quality value corresponding to each sample image output by the current first deep learning network model;
calculating a loss value based on each sample intrinsic quantity value, each labeled image quality value and a preset loss function;
judging whether the current first deep learning network model converges or not according to a loss value of a preset loss function; if so, taking the current first deep learning network model as a trained quality evaluation model; and if not, adjusting and updating the network parameters of the current first deep learning network model, and returning to the step of inputting each sample image in the training sample set into the current first deep learning network model and acquiring the sample quality value corresponding to each sample image output by the current first deep learning network model.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
An embodiment of the present invention further provides an electronic device, as shown in fig. 8, which includes a processor 801, a communication interface 802, a memory 803, and a communication bus 804, where the processor 801, the communication interface 802, and the memory 803 complete mutual communication through the communication bus 804,
a memory 803 for storing a computer program;
the processor 801 is configured to implement the following steps when executing the program stored in the memory 803:
obtaining an original image dataset containing a target object; the raw image dataset comprises a plurality of raw images;
extracting a target object area image containing only a target object from each original image in the original image data set;
inputting each target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
and taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image to form a target image data set containing the target object.
Therefore, by applying the embodiment of the invention, the target object area image only containing the target object is extracted from each original image in the original image data set, the influence of the image area where the non-target object is located on the image quality evaluation is reduced, the target object area image with the target quality value exceeding the preset quality value threshold is further obtained based on the quality evaluation model, the target image data set is formed, and the obtained target image data set is high in quality.
The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.
The communication interface is used for communication between the electronic equipment and other equipment.
The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component.
In a further embodiment of the present invention, a computer-readable storage medium is also provided, in which a computer program is stored, which, when being executed by a processor, carries out the steps of any of the above-mentioned image dataset acquisition methods.
In a further embodiment of the present invention, there is also provided a computer program product containing instructions which, when run on a computer, cause the computer to perform the method for acquiring an image data set according to any of the above embodiments.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for embodiments such as the apparatus, the electronic device, the computer-readable storage medium, and the computer program product, since they are substantially similar to the method embodiments, the description is simple, and for relevant points, reference may be made to part of the description of the method embodiments.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.
Claims (10)
1. A method of acquiring an image dataset, the method comprising:
obtaining an original image dataset containing a target object; the raw image dataset comprises a plurality of raw images;
extracting a target object area image containing only a target object from each original image in the original image data set;
inputting each target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
and taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image to form a target image data set containing the target object.
2. The method according to claim 1, wherein before the step of inputting each target object area image into a quality evaluation model and obtaining the target quality value of each target object area image output by the quality evaluation model, the method further comprises:
carrying out image transformation processing on each target object area image to obtain a corresponding target transformation image; the image transformation processing is used for increasing the number of target object area images and/or increasing the resolution;
the step of inputting the target object area images into a quality evaluation model and acquiring target quality values of the target object area images output by the quality evaluation model comprises the following steps:
and inputting the obtained target transformation images and the target object area images into a quality evaluation model as target object area images, and obtaining target quality values of the target object area images output by the quality evaluation model.
3. The method of claim 1, wherein the step of extracting a target object region image containing only a target object from each original image in the original image data set comprises:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as target object area images; the image recognition network model is as follows: and training a preset second deep learning network model by using an original sample image containing the target object and an image of a region where the target object is located in the original sample image as a labeled sample image in advance.
4. The method of claim 1, wherein the step of extracting a target object region image containing only a target object from each original image in the original image data set comprises:
respectively inputting each original image in the original image data set to a pre-trained image recognition network model, and acquiring a plurality of images output by the image recognition network model as images to be determined; the image recognition network model is as follows: the method comprises the steps that an original sample image containing a target object and an image of a region where the target object is located in the original sample image are used as marked sample images in advance, and a preset third deep learning network model is trained to obtain the marked sample images;
and acquiring an image with the resolution being greater than a preset resolution threshold and/or the noise intensity value being less than a preset noise intensity threshold in the image to be determined as a target object area image.
5. The method according to claim 4, wherein after the step of acquiring, as the target object region image, the image with the resolution greater than the preset resolution threshold and/or the noise intensity value less than the preset noise intensity threshold in the image to be determined, the method further comprises:
acquiring the definition of the residual images of the target object region removed from the plurality of images to be determined;
taking the residual image with the definition greater than a preset definition threshold value as a supplementary target object area image;
the step of inputting the target object area images into a quality evaluation model and acquiring target quality values of the target object area images output by the quality evaluation model comprises the following steps:
and inputting each target object area image and each supplemented target object area image into a quality evaluation model, and acquiring a target quality value of each target object area image output by the quality evaluation model.
6. The method according to claim 2, wherein the step of performing image transformation processing on each target object region image to obtain a corresponding target transformation image comprises:
performing multiple data enhancement processing on each target object region image by using a preset image data enhancement algorithm to obtain multiple target transformation images; and/or the presence of a gas in the gas,
based on an image transformation network model, enhancing or suppressing different characteristics in the target object area image, and obtaining a target transformation image corresponding to the target object area image; and/or the presence of a gas in the gas,
and inputting each target object area image into a preset super-resolution network model, and obtaining a target transformation image which is output by the preset super-resolution network model and has higher resolution than the target object area image.
7. The method of claim 1, wherein the training process of the quality assessment model comprises:
acquiring a training sample set; the training sample set comprises a plurality of training samples; wherein each training sample comprises: sample images and image quality values marked by the sample images;
inputting each sample image in the training sample set into a current first deep learning network model, and acquiring a sample quality value corresponding to each sample image output by the current first deep learning network model;
calculating a loss value based on each sample intrinsic quantity value, each labeled image quality value and a preset loss function;
judging whether the current first deep learning network model converges or not according to a loss value of a preset loss function; if so, taking the current first deep learning network model as a trained quality evaluation model; and if not, adjusting and updating the network parameters of the current first deep learning network model, and returning to the step of inputting each sample image in the training sample set into the current first deep learning network model and acquiring the sample quality value corresponding to each sample image output by the current first deep learning network model.
8. An apparatus for acquiring an image dataset, the apparatus comprising:
an original image data set obtaining unit for obtaining an original image data set containing a target object; the raw image dataset comprises a plurality of raw images;
a target object region image extracting unit configured to extract a target object region image containing only a target object from each original image in the original image data set;
a target quality value acquisition unit, configured to input each target object area image to a quality evaluation model, and acquire a target quality value of each target object area image output by the quality evaluation model; the quality evaluation model is as follows: training a preset first deep learning network model by using the sample image and the marked image quality value thereof to obtain the sample image;
and the target image data set forming unit is used for forming a target image data set containing a target object by taking the target object area image with the target quality value exceeding a preset quality value threshold as a target image.
9. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor and the communication interface are used for realizing mutual communication by the memory through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any of claims 1 to 7 when executing a program stored in the memory.
10. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, which computer program, when being executed by a processor, carries out the method steps of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010239580.5A CN111340140A (en) | 2020-03-30 | 2020-03-30 | Image data set acquisition method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010239580.5A CN111340140A (en) | 2020-03-30 | 2020-03-30 | Image data set acquisition method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111340140A true CN111340140A (en) | 2020-06-26 |
Family
ID=71184388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010239580.5A Pending CN111340140A (en) | 2020-03-30 | 2020-03-30 | Image data set acquisition method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111340140A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112270377A (en) * | 2020-11-11 | 2021-01-26 | 北京百度网讯科技有限公司 | Target image extraction method, neural network training method and device |
| WO2022156763A1 (en) * | 2021-01-25 | 2022-07-28 | 华为技术有限公司 | Target object detection method and device thereof |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198130A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Image processing method, device, storage medium and electronic equipment |
| CN108234870A (en) * | 2017-12-27 | 2018-06-29 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108664840A (en) * | 2017-03-27 | 2018-10-16 | 北京三星通信技术研究有限公司 | Image-recognizing method and device |
| WO2019104705A1 (en) * | 2017-12-01 | 2019-06-06 | 华为技术有限公司 | Image processing method and device |
| CN110163235A (en) * | 2018-10-11 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Training, image enchancing method, device and the storage medium of image enhancement model |
| CN110287791A (en) * | 2019-05-23 | 2019-09-27 | 罗普特科技集团股份有限公司 | A kind of screening technique and system for face picture |
| CN110782445A (en) * | 2019-10-25 | 2020-02-11 | 北京华捷艾米科技有限公司 | No-reference image quality evaluation method and system |
-
2020
- 2020-03-30 CN CN202010239580.5A patent/CN111340140A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664840A (en) * | 2017-03-27 | 2018-10-16 | 北京三星通信技术研究有限公司 | Image-recognizing method and device |
| WO2019104705A1 (en) * | 2017-12-01 | 2019-06-06 | 华为技术有限公司 | Image processing method and device |
| CN108234870A (en) * | 2017-12-27 | 2018-06-29 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108198130A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Image processing method, device, storage medium and electronic equipment |
| CN110163235A (en) * | 2018-10-11 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Training, image enchancing method, device and the storage medium of image enhancement model |
| CN110287791A (en) * | 2019-05-23 | 2019-09-27 | 罗普特科技集团股份有限公司 | A kind of screening technique and system for face picture |
| CN110782445A (en) * | 2019-10-25 | 2020-02-11 | 北京华捷艾米科技有限公司 | No-reference image quality evaluation method and system |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112270377A (en) * | 2020-11-11 | 2021-01-26 | 北京百度网讯科技有限公司 | Target image extraction method, neural network training method and device |
| CN112270377B (en) * | 2020-11-11 | 2024-03-15 | 北京百度网讯科技有限公司 | Target image extraction method, neural network training method and device |
| WO2022156763A1 (en) * | 2021-01-25 | 2022-07-28 | 华为技术有限公司 | Target object detection method and device thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110569899B (en) | Dam face defect classification model training method and device | |
| WO2020239015A1 (en) | Image recognition method and apparatus, image classification method and apparatus, electronic device, and storage medium | |
| CN108229526B (en) | Network training method, network training device, image processing method, image processing device, storage medium and electronic equipment | |
| CN112052787B (en) | Target detection method and device based on artificial intelligence and electronic equipment | |
| CN107784654B (en) | Image segmentation method and device and full convolution network system | |
| CN111080527B (en) | Image super-resolution method and device, electronic equipment and storage medium | |
| CN111080528B (en) | Image super-resolution and model training method and device, electronic equipment and medium | |
| CN113498528A (en) | Image defect determining method and device, electronic equipment and storage medium | |
| CN111476863B (en) | Method and device for coloring black-and-white cartoon, electronic equipment and storage medium | |
| CN110909663B (en) | Human body key point identification method and device and electronic equipment | |
| CN111462069B (en) | Training method and device for target object detection model, electronic equipment and storage medium | |
| WO2021082819A1 (en) | Image generation method and apparatus, and electronic device | |
| CN111340140A (en) | Image data set acquisition method and device, electronic equipment and storage medium | |
| CN111325067B (en) | Illegal video identification method and device and electronic equipment | |
| CN110909664A (en) | Human body key point identification method and device and electronic equipment | |
| CN110909665A (en) | Multitask image processing method and device, electronic equipment and storage medium | |
| CN111145202B (en) | Model generation method, image processing method, device, equipment and storage medium | |
| CN113033715B (en) | Target detection model training method and target vehicle detection information generation method | |
| CN113014928B (en) | Compensation frame generation method and device | |
| CN110852242A (en) | Watermark identification method, device, equipment and storage medium based on multi-scale network | |
| CN116958919A (en) | Target detection method, target detection device, computer readable medium and electronic equipment | |
| CN114119953A (en) | Method for quickly positioning and correcting license plate, storage medium and equipment | |
| CN114419086A (en) | Edge extraction method and device, electronic equipment and storage medium | |
| CN113128530B (en) | Data classification method and device | |
| CN117710365B (en) | Processing method and device for defective pipeline image and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |