CN112862702B

CN112862702B - Image enhancement method, device, equipment and storage medium

Info

Publication number: CN112862702B
Application number: CN202110064519.6A
Authority: CN
Inventors: 李佳琳; 王健宗; 瞿晓阳
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2021-01-18
Filing date: 2021-01-18
Publication date: 2023-10-31
Anticipated expiration: 2041-01-18
Also published as: WO2022151589A1; CN112862702A

Abstract

The invention relates to the field of big data and discloses an image enhancement method, an image enhancement device, image enhancement equipment and a storage medium. The method comprises the following steps: the preset image recognition model recognizes the scene image data acquired from the traffic accident scene and judges whether the scene image data meets the preset image standard or not; if the on-site image data does not meet the preset image standard, prompting the user to shoot again; if the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene; and inputting the target accident vehicle image into a preset image enhancement model for image enhancement, and generating a target enhancement image. According to the scheme, the conversion from low quality to high quality images can be completed only by learning the mapping function of the high quality images, the accuracy and processing timeliness of the vehicle insurance claim settlement are improved, and the technical problem of low claim settlement efficiency is solved.

Description

Image enhancement method, device, equipment and storage medium

Technical Field

The present invention relates to the field of big data, and in particular, to an image enhancement method, apparatus, device, and storage medium.

Background

Along with popularization of personal mobile phones and advancement of science and technology, imaging processing flows in traditional car insurance claim settlement flows generally need professional teams, use professional photographic equipment, and shoot and acquire image data of each area according to a complete acquisition flow. The number of private cars and various small accidents are increased, and the car insurance claims are subjected to larger pressure, so that the number of people and the processing efficiency can not meet the current demands, and the self-service, online material submitting, rapid claim settlement and even full-automatic claim settlement in recent years are marked selling points of internet insurance claim settlement services.

However, due to the quality of the image or video information collected in the step of photographing and uploading the automobile insurance claim material, the accuracy of the subsequent step is directly determined, the common user often does not have the operation skill of the professional, the used equipment is a camera of the common mobile phone, the photographing condition of the place cannot be determined when the automobile insurance is lost, and therefore, the photographed image or video may have the problems of wrong photographing target, unreasonable photographing angle, blurred imaging, insufficient exposure or strong light reflection and the like, and compared with the image phase difference collected by the professional, the problems increase the difficulty of OCR recognition or manual auditing in the later period and the processing efficiency cannot be improved.

Disclosure of Invention

The invention mainly aims to improve the accuracy and processing timeliness of vehicle insurance claim settlement and solve the technical problem of low claim settlement efficiency.

The first aspect of the present invention provides an image enhancement method, comprising:

acquiring scene image data of a traffic accident scene;

identifying the field image data through a preset image identification model, and judging whether the field image data meets a preset image standard or not;

if not, prompting the user to shoot again;

if yes, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

and inputting the target accident vehicle image into a preset image enhancement model, and enhancing the target accident vehicle image to generate a target enhancement image.

Optionally, in a first implementation manner of the first aspect of the present invention, before the acquiring the field image data of the traffic accident scene, the method further includes:

acquiring historical vehicle images of an accident scene from a preset database;

constructing a training sample data set by taking the historical vehicle image as a training sample;

inputting the training sample data set into a preset image enhancement generator, and training the image enhancement generator to obtain a target image enhancement model.

Optionally, in a second implementation manner of the first aspect of the present invention, before the identifying, by the preset image identifying model, the live image data, and before determining whether the live image data meets a preset image standard, the method further includes:

acquiring a plurality of training pictures from a preset training picture database, and inputting the training pictures into a preset neural network model to obtain a prediction type label of the training pictures;

generating a real class label of the training picture according to the class of the training picture;

generating a first loss function of the training picture according to the prediction type label of the training picture and the real type label of the training picture;

and updating parameters in the neural network model according to the first loss function to obtain a target picture identification model.

Optionally, in a third implementation manner of the first aspect of the present invention, inputting the training sample data set into a preset image enhancement generator, and training the image enhancement generator to obtain the target image enhancement model includes:

performing image enhancement on the training sample data set to obtain a preset number of target original images and corresponding enhanced images;

Inputting the target original image into a CNN image enhancement network in a preset image enhancement generator to obtain an output image corresponding to the target original image;

determining a second loss function between the output image and the enhanced image;

and performing iterative training on the CNN image enhancement network in the image enhancement generator based on the second loss function until the CNN image enhancement network converges to obtain a target image enhancement model.

Optionally, in a fourth implementation manner of the first aspect of the present invention, the inputting the target accident vehicle image into a preset image enhancement model, performing image enhancement on the target accident vehicle image, and generating the target enhancement image includes:

performing downsampling processing on the target accident vehicle image to obtain a downsampled image;

inputting the downsampled image into the target image enhancement model to obtain image enhancement data corresponding to the target accident vehicle image;

determining a matching point of each pixel point in the target accident vehicle image in the downsampled image;

determining target enhancement parameters corresponding to the pixel points based on enhancement data corresponding to the matching points of the pixel points in the image enhancement data;

And adjusting pixel values of the pixel points based on the target enhancement parameters corresponding to the pixel points to obtain a target enhancement image corresponding to the target accident vehicle image.

Optionally, in a fifth implementation manner of the first aspect of the present invention, after the inputting the target accident vehicle image into a preset image enhancement model, performing image enhancement on the target accident vehicle image, generating a target enhancement image, the method further includes:

acquiring owner identity information of a user and damage information of the accident vehicle, wherein the vehicle information comprises vehicle risk information of the accident vehicle;

receiving a claim settlement request uploaded by the user, and inputting the target enhanced image into a trained convolutional neural network model for prediction to obtain the claim settlement probability of the target enhanced image;

judging whether the claim settling probability of the target enhanced image is larger than a preset threshold value;

if the claim settlement probability of the target enhanced picture is larger than a preset threshold value, determining that the target enhanced image can settle the claim;

and carrying out claim settlement on the accident vehicle based on the owner identity information of the user and the identification information of the accident vehicle.

A second aspect of the present invention provides an image enhancement apparatus comprising:

The first acquisition module is used for acquiring scene image data of a traffic accident scene;

the first judging module is used for identifying the field image data through a preset image identification model and judging whether the field image data meets a preset image standard or not;

the prompting module is used for prompting a user to shoot again when the on-site image data does not meet the preset image standard;

the shooting module is used for entering a next shooting task when the scene image data meets a preset image standard to obtain a target accident vehicle image of the traffic accident scene;

the image enhancement module is used for inputting the target accident vehicle image into a preset image enhancement model, and carrying out image enhancement on the target accident vehicle image to generate a target enhancement image.

Optionally, in a first implementation manner of the second aspect of the present invention, the image enhancement device further includes:

the second acquisition module is used for acquiring historical vehicle images of the accident scene from a preset database;

the construction module is used for constructing a training sample data set by taking the historical vehicle image as a training sample;

the training module is used for inputting the training sample data set into a preset image enhancement generator, and training the image enhancement generator to obtain a target image enhancement model.

Optionally, in a second implementation manner of the second aspect of the present invention, the image enhancement device further includes:

the input module is used for acquiring a plurality of training pictures from a preset training picture database, presetting a neural network model for the training pictures, and obtaining a prediction type label of the training pictures;

the generation module is used for generating a real class label of the training picture according to the class of the training picture; generating a first loss function of the training picture according to the prediction type label of the training picture and the real type label of the training picture;

and the updating module is used for updating parameters in the neural network model according to the first loss function to obtain a target picture identification model.

Optionally, in a third implementation manner of the second aspect of the present invention, the training module includes:

the image enhancement unit is used for carrying out image enhancement on the training sample data set to obtain a preset number of target original images and corresponding enhanced images;

the input unit is used for inputting the target original image into a CNN image enhancement network in a preset image enhancement generator to obtain an output image corresponding to the target original image;

A determining unit configured to determine a second loss function between the output image and the enhanced image;

and the iterative training unit is used for carrying out the CNN image enhancement network in the image enhancement generator based on the second loss function until the CNN image enhancement network converges to obtain a target image enhancement model.

Optionally, in a fourth implementation manner of the second aspect of the present invention, the image enhancement module is specifically configured to:

Optionally, in a fifth implementation manner of the second aspect of the present invention, the image enhancement device further includes:

The third acquisition module is used for acquiring the owner identity information of the user and the damage information of the accident vehicle, wherein the vehicle information comprises the vehicle risk information of the accident vehicle;

the prediction module is used for receiving the claim settlement request uploaded by the user, inputting the target enhanced image into the trained convolutional neural network model for prediction, and obtaining the claim settlement probability of the target enhanced image;

the second judging module is used for judging whether the claim settling probability of the target enhanced image is larger than a preset threshold value or not;

the determining module is used for determining that the target enhanced image can be subjected to claim settlement when the claim settlement probability of the target enhanced image is larger than a preset threshold value;

and the claim settlement module is used for settling the accident vehicle based on the owner identity information of the user and the identification information of the accident vehicle.

A third aspect of the present invention provides an image enhancement apparatus comprising: a memory and at least one processor, the memory having instructions stored therein, the memory and the at least one processor being interconnected by a line;

the at least one processor invokes the instructions in the memory to cause the image enhancement device to perform the image enhancement method described above.

A fourth aspect of the present invention provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the above-described image enhancement method.

In the technical scheme provided by the invention, the scene image data acquired from the traffic accident scene is identified through the preset image identification model, and whether the scene image data meets the preset image standard is judged; if the on-site image data does not meet the preset image standard, prompting the user to shoot again; if the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene; and inputting the target accident vehicle image into a preset image enhancement model for image enhancement, and generating a target enhancement image. According to the scheme, the conversion from low quality to high quality images can be completed only by learning the mapping function of the high quality images, the accuracy and processing timeliness of the vehicle insurance claim settlement are improved, and the technical problem of low claim settlement efficiency is solved.

Drawings

FIG. 1 is a schematic diagram of a first embodiment of an image enhancement method of the present invention;

FIG. 2 is a schematic diagram of a second embodiment of the image enhancement method of the present invention;

FIG. 3 is a schematic diagram of a third embodiment of the image enhancement method of the present invention;

FIG. 4 is a schematic diagram of a fourth embodiment of the image enhancement method of the present invention;

FIG. 5 is a schematic diagram of a fifth embodiment of the image enhancement method of the present invention;

FIG. 6 is a schematic diagram of a first embodiment of an image enhancement device of the present invention;

FIG. 7 is a schematic diagram of a second embodiment of an image enhancement device of the present invention;

fig. 8 is a schematic diagram of an embodiment of the image enhancement device of the present invention.

Detailed Description

The embodiment of the invention provides an image enhancement method, an image enhancement device, image enhancement equipment and a storage medium, wherein in the technical scheme of the invention, an image recognition model is preset to recognize field image data acquired from a traffic accident scene at first, and whether the field image data meets preset image standards is judged; if the on-site image data does not meet the preset image standard, prompting the user to shoot again; if the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene; and inputting the target accident vehicle image into a preset image enhancement model for image enhancement, and generating a target enhancement image. According to the scheme, the conversion from low quality to high quality images can be completed only by learning the mapping function of the high quality images, the accuracy and processing timeliness of the vehicle insurance claim settlement are improved, and the technical problem of low claim settlement efficiency is solved.

The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For easy understanding, the following describes a specific flow of an embodiment of the present invention, referring to fig. 1, and a first embodiment of an image enhancement method in an embodiment of the present invention includes:

101. acquiring scene image data of a traffic accident scene;

in this embodiment, in order to ensure that sufficient information is acquired, the user needs to take a 360-degree image of the accident vehicle and an image of the critical damaged portion at the traffic accident site. For example, a user needs to take images of corresponding vehicle parts including a frame number, a damaged part, a car whole-body photo, and the like through a personal mobile phone on site according to directions.

102. Identifying the field image data through a preset image identification model, and judging whether the field image data meets preset image standards or not;

in this embodiment, since the quality of the image or video information collected in the step of capturing and uploading the vehicle insurance claim material will directly determine the accuracy of the subsequent step, the ordinary user often does not have the operation skill of the professional, the used device is the camera of the ordinary mobile phone, and the capturing condition of the location cannot be determined when the vehicle insurance is lost, so that the captured image or video may have the problems of wrong capturing target, unreasonable capturing angle, blurred imaging, insufficient exposure or strong light reflection, and the like, so that the on-site image data of the accident vehicle captured by the user needs to be identified through the preset image identification model, whether the image is the corresponding photo required in the guide is determined through the identification model, and whether the image meets the preset standard is rapidly determined through the schemes such as edge detection.

103. When the field image data does not meet the preset image standard, prompting the user to shoot again;

in this embodiment, whether the photographed image meets the corresponding photographing standard is locally determined, whether the photographed image is the corresponding photograph required in the guide is determined through the recognition model, and whether the blurring condition exists is rapidly determined through schemes such as edge detection. And if the standard is not met, the user is prompted to shoot again.

104. When the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

in this embodiment, if the picture shot by the user on site meets the preset requirement, the next shooting task is performed, and an image with higher quality including brightness, contrast, white balance, latitude, noise, details, color transition and the like, that is, the target enhanced image is obtained.

105. Inputting the target accident vehicle image into a preset image enhancement model, and performing image enhancement on the target accident vehicle image to generate a target enhancement image.

In this embodiment, a low-quality target accident vehicle image shot by a mobile phone is input into a target image enhancement model to complete comprehensive promotion in terms of brightness, contrast, white balance, latitude, noise, details, color transition and the like, so as to generate an enhanced picture which is based on the original picture content and achieves a visual effect similar to that shot by a professional digital camera.

In this embodiment, the method of image enhancement is to attach some information or transformation data to the original image by a certain means, so as to selectively highlight the interesting features in the image or suppress (mask) some unwanted features in the image, so as to match the image with the visual response characteristics. In the image enhancement process, the reason of image degradation is not analyzed, and the processed image does not necessarily approach the original image. The image enhancement technology can be divided into two major categories, namely an algorithm based on a space domain and an algorithm based on a frequency domain according to different spaces in which enhancement processing processes are located.

In the embodiment of the invention, the on-site image data acquired from the traffic accident scene is identified through the preset image identification model, and whether the on-site image data meets the preset image standard is judged; if the on-site image data does not meet the preset image standard, prompting the user to shoot again; if the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene; and inputting the target accident vehicle image into a preset image enhancement model for image enhancement, and generating a target enhancement image. According to the scheme, the conversion from low quality to high quality images can be completed only by learning the mapping function of the high quality images, the accuracy and processing timeliness of the vehicle insurance claim settlement are improved, and the technical problem of low claim settlement efficiency is solved.

Referring to fig. 2, a second embodiment of an image enhancement method according to an embodiment of the present invention includes:

201. acquiring historical vehicle images of an accident scene from a preset database;

in this embodiment, after a traffic accident, at least one target image of an accident vehicle at the traffic accident scene is acquired, and the target images can reflect specific situations of the accident scene, such as license plate numbers, damaged parts and damaged degrees of the accident vehicle.

202. Constructing a training sample data set by taking the historical vehicle image as a training sample;

in this embodiment, a plurality of original image samples in various vehicle insurance claim scenes may be obtained from a database, where the original image samples may be images taken by a user from an event site, or may be images taken by a company professional with a professional camera, or the like. For example, the photo shot by the single lens reflex has few noise points and rich detail information. And then, responding to user operation, carrying out image enhancement on the image parameters of each original image sample, and generating an enhanced image sample after image enhancement. For example, each original image sample can be adjusted by a professional special effect designer to obtain an enhanced image sample after image enhancement, so that the enhanced image sample has better brightness and saturation.

203. Inputting the training sample data set into a preset image enhancement generator, and training the image enhancement generator to obtain a target image enhancement model;

in this embodiment, image enhancement refers to enhancing useful information in an image, which may be a distorting process that aims to improve the visual effect of an image for a given image application. The method has the advantages that the integral or local characteristics of the image are purposefully emphasized, the original unclear image is changed into clear or some interesting features are emphasized, the differences among different object features in the image are enlarged, the uninteresting features are restrained, the image quality is improved, the information quantity is enriched, the image interpretation and recognition effects are enhanced, and the needs of some special analysis are met.

In this embodiment, the training sample data is input into a preset image enhancement generator to train the image enhancement generator, and the specific implementation steps include: the original low-quality image x is input, an image y with improvement in aspects of brightness, exposure and the like is generated through the generator, the texture and color discriminators calculate texture and color loss functions by comparing the image y with a picture z shot by a professional camera, and the generator is trained through secondary feedback. Meanwhile, a low-quality picture y 'is reversely generated by applying a generator G' to the image y, and the generation control of the generator G is achieved by calculating the content loss with the original picture x.

204. Acquiring scene image data of a traffic accident scene;

205. identifying the field image data through a preset image identification model, and judging whether the field image data meets preset image standards or not;

206. when the field image data does not meet the preset image standard, prompting the user to shoot again;

207. when the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

208. inputting the target accident vehicle image into a preset image enhancement model, and performing image enhancement on the target accident vehicle image to generate a target enhancement image.

Steps 204 to 208 in this embodiment are similar to steps 101 to 105 in the first embodiment, and will not be described here.

Referring to fig. 3, a third embodiment of an image enhancement method according to an embodiment of the present invention includes:

301. acquiring historical vehicle images of an accident scene from a preset database;

302. constructing a training sample data set by taking the historical vehicle image as a training sample;

303. Inputting the training sample data set into a preset image enhancement generator, and training the image enhancement generator to obtain a target image enhancement model;

304. acquiring scene image data of a traffic accident scene;

305. acquiring a plurality of training pictures from a preset training picture database, and inputting the training pictures into a preset neural network model to obtain a prediction type label of the training pictures;

in the embodiment, a picture of an accident vehicle shot at a past vehicle accident scene is obtained from a preset database and used as a picture to be identified, and a prediction type label of the picture to be identified is obtained; the pre-trained picture recognition model adopts a convolutional neural network model, and the convolutional neural network model comprises at least one nuclear pooling layer for increasing the channel dimension from one dimension to multiple dimensions.

In this embodiment, parameters in the picture recognition model are continuously changed during the process of training the picture recognition model, that is, the picture recognition model is continuously changed. And in each training, the adopted picture recognition model is the picture recognition model updated in the last training. The parameters of the picture recognition model in this embodiment are preset initial values when the picture recognition model is used for the first time. The dimension of the prediction category label is determined during training, for example, if a plurality of training pictures in the training picture database include n classified pictures, the dimension of the prediction category label output by the picture identification model may be set to be n.

306. Generating a real class label of the training picture according to the class of the training picture;

in this embodiment, since the category of the training picture is known, the real category label of the category of the training picture may be generated according to the category of the training picture. Similarly, the real class label of the training picture is also in the form of a vector, in which only the value of the element at the position corresponding to the class of the training picture is 1, and the rest positions are all 0.

307. Generating a first loss function of the training picture according to the prediction type label of the training picture and the real type label of the training picture;

in this embodiment, according to the obtained prediction class label of the training picture and the obtained real class label of the training picture, the difference between the prediction class label of the training picture and the real class label of the training picture can be known, and a loss function of the training picture is generated according to the difference. In the process of training the image recognition model, the value of the loss function is gradually reduced, so that the trained image recognition model is more accurate. In this embodiment, there are more than two loss functions, feature differences of the tag are calculated respectively, a plurality of feature differences calculated according to the corresponding loss functions are ordered, the loss function corresponding to the feature difference with the largest order is used as the final loss function, and feedback adjustment weights are performed on the model.

In this embodiment, because the loss function represents a value that decreases toward the gradient direction, the larger the loss function value, the more inaccurate the prediction type label that represents the picture recognition model and that is output after the input picture is processed, in this embodiment, the gradient of the final point of the loss function is made to be 0 by continuously adjusting the loss function. The smaller the value of the loss function, the more accurate the processing result representing the picture recognition model. That is, the derivative of the loss function of the training picture to each parameter in the picture recognition model is calculated, and each parameter in the picture recognition model is updated according to each parameter in the picture recognition model and the derivative of the loss function corresponding to the training picture to each parameter, so that the picture recognition model after the training is determined.

The loss function (loss function), also called cost function, is a function that maps the value of a random event or its related random variable to a non-negative real number to represent the "risk" or "loss" of the random event. In application, the loss function is typically associated with an optimization problem as a learning criterion, i.e., solving and evaluating the model by minimizing the loss function. For example, in statistics and machine learning, are used for parameter estimation of models (parametric estimation). In this embodiment, the loss function is pre-constructed to calculate the feature difference between the real feature and the predicted feature.

308. Updating parameters in the neural network model according to the first loss function to obtain a target picture identification model;

in this embodiment, according to each parameter in the image recognition model and the derivative of the loss function of the training image on each parameter, each parameter in the image recognition model is updated to obtain the target image recognition model.

Specifically, since the picture recognition model includes a plurality of parameters, when the input training picture is determined, the loss function can also be regarded as a function of the parameters in the picture recognition model, and the derivative of the loss function of the training picture on each parameter in the picture recognition model can be calculated; for each parameter in the picture recognition model, the parameter in the picture recognition model can be updated to be the parameter minus the derivative of the loss function with respect to the parameter when updated, i.e., the picture recognition model is updated every training.

Because the loss function represents a value decreasing toward the gradient direction, the larger the loss function value is, the more inaccurate the prediction type label which represents the picture recognition model and is output after the input picture is processed is, in this embodiment, the gradient of the final point of the loss function is 0 by continuously adjusting the loss function. The smaller the value of the loss function, the more accurate the processing result representing the picture recognition model. In the technical solution of the above embodiment, when updating the picture recognition model, the picture recognition model is updated in a direction gradually decreasing toward the loss function. Therefore, through continuous training, the loss function is continuously optimized, and parameters of the picture identification model are continuously updated, so that the picture identification model is continuously updated, and the finally obtained picture identification model can very accurately identify the picture.

309. Identifying the field image data through a preset image identification model, and judging whether the field image data meets preset image standards or not;

310. when the field image data does not meet the preset image standard, prompting the user to shoot again;

311. when the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

312. inputting the target accident vehicle image into a preset image enhancement model, and performing image enhancement on the target accident vehicle image to generate a target enhancement image.

Steps 301-303, 309-312 in this embodiment are similar to steps 201-203, 102-105 in the first embodiment, and will not be described here again.

Referring to fig. 4, a fourth embodiment of the image enhancement method according to the present invention includes:

401. acquiring historical vehicle images of an accident scene from a preset database;

402. constructing a training sample data set by taking the historical vehicle image as a training sample;

403. performing image enhancement on the training sample data set to obtain a preset number of target original images and corresponding enhanced images;

in this embodiment, a plurality of low-quality images are obtained from a preset database as training sample data, and these images are enhanced to obtain a preset number of original image samples and a corresponding preset number of enhanced image samples. Then, for each selected original image sample and corresponding preset number of enhanced image samples, randomly cropping the image with the target size at the same position in the original image sample and corresponding enhanced image sample to obtain the original image with the preset number of target sizes and corresponding enhanced image.

404. Inputting the target original image into a CNN image enhancement network in a preset image enhancement generator to obtain an output image corresponding to the target original image;

in this embodiment, the collected target original images are input into the CNN image enhancement network in the preset image enhancement generator, so that an output image corresponding to each target original image can be obtained.

405. Determining a second loss function between the output image and the enhanced image;

in this embodiment, the loss function is pre-constructed and used to calculate the feature difference between the real feature and the predicted feature, rank the feature differences calculated according to the corresponding loss function, take the loss function corresponding to the feature difference with the largest rank as the final loss function, and monitor the training of the CNN image enhancement network and update the network parameters based on the loss function value after obtaining the loss function value.

In this embodiment, the blurring process may be performed on the output image and the enhanced image by using a mean blurring method, and a mean error value of the blurred output image and the enhanced image may be calculated, to obtain a color_loss value. The blurring process is carried out on the output image and the enhanced image by utilizing a mean blurring method, so that the interference of high-frequency information is eliminated, and the network learns the color information more. In practical implementation, the training of end2end can be implemented by using a mean pooling layer in the CNN image enhancement network and setting stride to be 1, so that l2loss (i.e. Mean Square Error Loss) is performed on the output image and the enhancement image after the blurring process to obtain the color_loss value.

Further, performing error calculation on the feature images output by the output image and the enhanced image at the same layer through the CNN image enhancement network to obtain a vgg _loss value. Where vgg _loss is a semantic type loss, semantic information can be better generated. Illustratively, the present embodiment can use vgg as a network structure to generate different layer feature maps (feature maps) while initializing vgg the network with trained network parameters on the ImageNet dataset. And carrying out l2loss on the characteristic diagram output by the same layer through a vgg network on the output image and the enhanced image to obtain a vgg _loss value.

Further, respectively obtaining a gray level image of the output image and a gray level image of the enhanced image, and performing error calculation on the gray level image of the output image and the gray level image of the enhanced image to obtain a texture_loss value. texture_loss, namely texture loss, can be added with texture details on the basis of the texture_loss, and detail information is not lost while image enhancement is ensured.

406. Performing iterative training on the CNN image enhancement network in the image enhancement generator based on the second loss function until the CNN image enhancement network converges to obtain a target image enhancement model;

In this embodiment, after obtaining a loss function value, monitoring training of the CNN image enhancement network based on the loss function value and updating network parameters until the CNN image enhancement network converges to obtain a target image enhancement model;

in this embodiment, the manner of determining whether the CNN image enhancement network reaches the training convergence condition may be: judging whether the variation value of the loss value is smaller than a preset value, if the variation value of the loss value is smaller than the preset value, judging that the CNN image enhancement network reaches a training convergence condition, otherwise, judging that the CNN image enhancement network does not reach the training convergence condition. The preset value may be set according to actual requirements, for example, the preset value may be set to a value close to 0, or may also be set to 0, if the preset value is 0, when the loss value does not change, it is determined that the CNN image enhancement network reaches the training convergence condition, or else it is determined that the CNN image enhancement network does not reach the training convergence condition. And if the CNN image enhancement network does not reach the training convergence condition, repeating the steps to continue training.

When the CNN image enhancement network reaches a training convergence condition, updating network parameters of the CNN image enhancement network, so as to output a corresponding image enhancement model. The image enhancement model may perform image enhancement on an input image. In detail, after receiving the image to be processed, the image to be processed may be input into the image enhancement model to obtain an enhanced image corresponding to the image to be processed. For example, the user inputs the image shot by the mobile phone as the image to be processed into the trained image enhancement model, and then the corresponding enhancement image is obtained through layer-by-layer calculation of the network, and the corresponding enhancement image is the target enhancement image.

407. Acquiring scene image data of a traffic accident scene;

408. identifying the field image data through a preset image identification model, and judging whether the field image data meets preset image standards or not;

409. when the field image data does not meet the preset image standard, prompting the user to shoot again;

410. when the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

411. inputting the target accident vehicle image into a preset image enhancement model, and performing image enhancement on the target accident vehicle image to generate a target enhancement image.

Steps 407 to 411 in this embodiment are similar to steps 101 to 105 in the first embodiment, and will not be described here.

Referring to fig. 5, a fifth embodiment of an image enhancement method according to an embodiment of the present invention includes:

501. acquiring scene image data of a traffic accident scene;

502. identifying the field image data through a preset image identification model, and judging whether the field image data meets preset image standards or not;

503. when the on-site image data does not meet the preset image standard, prompting a user to shoot again;

504. when the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene;

505. performing downsampling processing on the target accident vehicle image to obtain a downsampled image;

in this embodiment, when an image enhancement process is required for a certain image, the image may be used as a target enhancement image to be enhanced. The target enhanced image may be a high resolution image. Moreover, the target enhanced image may be a single-channel image or a multi-channel image. Wherein the color space of the multi-channel image may be: RGB (Red, green, blue), YUV (luminence, chroma, brightness) or other color space, as the disclosure is not limited in this regard. The electronic equipment can shoot the target enhanced image through an internal or external camera device, can also communicate with other equipment, and can receive the target enhanced image sent by other equipment. The present disclosure is not limited in the manner in which the electronic device obtains the target enhanced image.

After the electronic device acquires the target enhanced image, the target enhanced image may be subjected to downsampling processing, thereby obtaining a downsampled image. The resulting downsampled image may further be input to a deep learning network, which performs image enhancement processing on the downsampled image.

In this embodiment, the target enhanced image may be subjected to a downsampling process by x times, so that the resolution, width, and height of the obtained downsampled image are each 1/x times that of the corresponding parameters of the target enhanced image.

506. Inputting the downsampled image into a target image enhancement model to obtain image enhancement data corresponding to the target image enhancement model;

in this embodiment, in order to perform image enhancement processing on the downsampled image, the deep learning network may be trained in advance according to the sample image and the sample enhancement image corresponding to the sample image, so as to obtain a trained deep learning network. Wherein the sample image and the corresponding sample enhancement image are training samples for training. The sampled image and corresponding sample enhancement image may be obtained in an existing training library as training samples. The sample enhanced image may be obtained by multi-exposure fusion, or may be obtained by other single-frame image enhancement methods, which is not limited in this disclosure.

In this embodiment, the structure of the deep learning network may be any existing deep learning network model. Specifically, during training, the number of training samples can be determined according to actual needs. Meanwhile, a reasonable loss function or objective function and a corresponding target value can be set to determine whether the deep learning network is trained well.

After the deep learning network is trained, network parameters of the deep learning network are determined. Because the deep learning network is trained according to the sample image and the corresponding sample enhancement image, the image enhancement data corresponding to the downsampled image can be obtained after the downsampled image is input into the trained deep learning network. The image enhancement data is data characterizing the degree of enhancement of the downsampled image relative to the downsampled image. The image enhancement data may take a variety of forms, such as: the image enhancement data may include mapping parameters for mapping each pixel point in the downsampled image to a corresponding pixel point, where the corresponding pixel point of any pixel point is the same pixel point as the pixel point in the downsampled image enhanced image. For another example, the image enhancement data may include downsampled image enhanced images. Moreover, it can be understood that the output result of the deep learning network is the same as the type of the image enhancement data corresponding to the downsampled image, that is, when the output result of the deep learning network is the mapping parameter, the image enhancement data corresponding to the downsampled image is the mapping parameter, and when the output result of the deep learning network is the enhanced image, the image enhancement data corresponding to the downsampled image is the image enhanced by the downsampled image.

507. Determining a matching point of each pixel point in the target accident vehicle image in the downsampled image;

in this embodiment, since the resolution of the downsampled image is lower than the resolution of the target enhanced image, the image enhanced data corresponding to the downsampled image does not correspond to the target enhanced image. In order to obtain an enhanced image corresponding to the target enhanced image, in this embodiment, a matching point of each pixel point in the target enhanced image in the downsampled image may be determined first, then, a target enhanced parameter corresponding to the pixel point is determined, and finally, a pixel value of the pixel point is adjusted by using the target enhanced parameter.

In this embodiment, for each pixel in the target enhanced image, a corresponding pixel in the downsampled image of the pixel is determined, and in a search area centered on the corresponding pixel and having a size of mxn, a pixel having the smallest absolute value of a difference between the pixel and the pixel value of the corresponding pixel is found, and the found pixel is used as a matching point of the pixel in the downsampled image.

508. Determining target enhancement parameters corresponding to the pixel points based on enhancement data corresponding to the matching points of the pixel points in the image enhancement data;

In this embodiment, after determining a matching point of each pixel point in the target enhanced image in the downsampled image, a target enhancement parameter corresponding to the pixel point may be determined based on enhancement data corresponding to the matching point, and a pixel value of the pixel point may be further adjusted based on the target enhancement parameter. Specifically, the image enhancement data are different in form, and the mode of determining the target enhancement parameter corresponding to each pixel point in the target enhancement image is also different. Two ways are given below for illustration.

Optionally, the image enhancement data may include: each pixel point in the downsampled image is mapped into a mapping parameter of a corresponding pixel point, and the corresponding pixel point of any pixel point is the pixel point with the same position as the pixel point in the image after the downsampled image is enhanced. The mapping of a pixel to a corresponding pixel is specifically: and adjusting the pixel value of one pixel point to the pixel value of the corresponding pixel point.

Correspondingly, for each pixel point in the target enhanced image, determining the target enhancement parameter corresponding to the pixel point based on the enhancement data corresponding to the matching point of the pixel point in the image enhancement data may include: for each pixel point in the target enhanced image, determining a target parameter corresponding to a matching point of the pixel point from the mapping parameters, and taking the determined target parameter as the target enhanced parameter corresponding to the pixel point.

509. Based on the target enhancement parameters corresponding to the pixel points, adjusting the pixel values of the pixel points to obtain a target enhancement image corresponding to the target accident vehicle image;

in this embodiment, for each pixel point in the target enhanced image, after determining the target enhanced parameter corresponding to the pixel point, the pixel value of the pixel point may be adjusted based on the target enhanced parameter. This adjustment process is a process of performing image enhancement processing on the target enhanced image.

510. Acquiring owner identity information of a user and damage information of an accident vehicle, wherein the vehicle information comprises vehicle risk information of the accident vehicle;

in this embodiment, the accident type, the relative position information of the accident vehicle, the first vehicle identification of the accident vehicle, the damaged position of the accident vehicle, and the damage degree of the accident vehicle are transmitted to the client.

The accident type is a precondition for judging the damage of the accident vehicle, different accident types can influence the damaged position and the damaged degree of the accident vehicle, and the accident types comprise: rear-end collisions, scratches, merging collisions, cornering collisions, etc. For rear-end collision accidents, the damaged positions of the accident vehicles are the front end and the rear end of the accident vehicles, most of the accident vehicles are damaged by collision, and the damage degree is relatively heavy compared with scratch accidents; for parallel collision and scraping accidents, the damaged positions of the accident vehicle are two sides of the accident vehicle, and most of the damaged positions are scraping injuries; for a corner crash accident, the damaged location of the accident vehicle is one side and the front end of the accident vehicle, and there is both crash damage and scratch damage. Therefore, it is necessary to determine the accident type from the accident image. And according to the owner identity information and the damage information of the accident vehicle, which are input by the user, the vehicle is subjected to claim settlement.

511. Receiving a claim settlement request uploaded by a user, and inputting the target enhanced image into a trained convolutional neural network model for prediction to obtain the claim settlement probability of the target enhanced image;

in this embodiment, the claim settlement request uploaded by the user includes a target enhanced image (i.e., the target enhanced image in this embodiment) and claim settlement information, where the target enhanced image is a picture taken by the user at a damaged portion of the vehicle where a vehicle accident occurs, for example, a picture of a rear-end collision of the vehicle, a picture of a side-turning of the vehicle, and a picture of a scratch of the vehicle. The claim information refers to information related to claim uploading to an insurance company, such as an automobile model number, a damage cause, a policy number, etc., by a user after a vehicle accident. After a vehicle accident, a user fills in basic vehicle information, accident cause, takes a picture of a damaged vehicle on site, and uploads a target enhanced image and claim information to a server of an insurance company to generate a claim settlement request. And after receiving the claim settlement request uploaded by the user, calling a trained convolutional neural network model, and inputting the target enhanced image into the convolutional neural network model for prediction to obtain the claim settlement probability of the target enhanced image, namely the probability of whether the target enhanced image should be subjected to claim settlement.

512. Judging whether the claim settling probability of the target enhanced image is larger than a preset threshold value;

in this embodiment, a classifier SVM (support vector machine) is used for classifying, which is a classification model for classifying the target enhanced image, wherein one class is classified as belonging to the claim, and the other class is classified as not belonging to the claim. After the one-dimensional feature vector is obtained, the one-dimensional feature vector is input into a classifier, and the classifier maps the one-dimensional feature vector to a numerical value in a range from 0 to 1, wherein the numerical value is the claim settlement probability.

513. When the claim settlement probability of the target enhanced picture is larger than a preset threshold value, determining that the target enhanced image can settle the claim;

in this embodiment, after obtaining the claim probability of the target enhanced image, the claim probability is compared with a preset threshold value, and the preset threshold value is 0.7, which can be understood that any other value is also possible. When the claim settlement probability corresponding to the target enhanced image is larger than a preset threshold value, the target enhanced image is very close to the target accident vehicle image which has been subjected to claim settlement, and the target enhanced image is judged to belong to the claimable claim. And when the claim settlement probability corresponding to the target enhanced image is smaller than a preset threshold value, indicating that the target enhanced image has the risk of cheating protection or fraud, and judging that the target enhanced image does not belong to the claimable claim. For example, when the probability of claim settlement is 0.8, then the probability of claim settlement is greater than a preset threshold, and then it is determined that the target enhanced image is claimable.

514. And (5) carrying out claim settlement on the accident vehicle based on the owner identity information of the user and the identification information of the accident vehicle.

In this embodiment, all the claims cases are screened from the preset database, the claims cases with the claims are selected, and the claims information of the screened claims cases is obtained, wherein the claims cases with the claims are marked after the claims are finished, so that the claims cases with the marks can be screened and obtained by selecting the claims cases with the marks; and then matching the claim information of the screened claim case with the claim information uploaded by the user according to preset rules, wherein the preset rules are matched according to the vehicle model, the vehicle age and the damaged position, for example, if the vehicle model uploaded by the user is A-type, the vehicle age is 2 years and the damaged position is the tail of the vehicle, the claim information matched one by one is searched from the claim information correspondingly, the amount of the claim which is already in the matched claim information is taken as the estimated amount of the claim, and the estimated amount of the claim is sent to the user.

Steps 501 to 504 in this embodiment are similar to steps 101 to 104 in the first embodiment, and will not be described here again.

In the embodiment of the invention, the scene image data acquired from the traffic accident scene is identified through the preset image identification model, and whether the scene image data meets the preset image standard is judged; if the on-site image data does not meet the preset image standard, prompting the user to shoot again; if the scene image data meets the preset image standard, entering the next shooting task to obtain a target accident vehicle image of the traffic accident scene; and inputting the target accident vehicle image into a preset image enhancement model for image enhancement, and generating a target enhancement image. According to the scheme, the conversion from low quality to high quality images can be completed only by learning the mapping function of the high quality images, the accuracy and processing timeliness of the vehicle insurance claim settlement are improved, and the technical problem of low claim settlement efficiency is solved.

The image enhancement method in the embodiment of the present invention is described above, and the image enhancement apparatus in the embodiment of the present invention is described below, referring to fig. 6, where a first embodiment of the image enhancement apparatus in the embodiment of the present invention includes:

a first obtaining module 601, configured to obtain field image data of a traffic accident scene;

a first judging module 602, configured to identify the live image data through a preset image identification model, and judge whether the live image data meets a preset image standard;

a prompting module 603, configured to prompt a user to re-shoot when the live image data does not meet a preset image standard;

the shooting module 604 is configured to enter a next shooting task to obtain a target accident vehicle image of the traffic accident scene when the scene image data meets a preset image standard;

the image enhancement module 605 is configured to input the target accident vehicle image into a preset image enhancement model, perform image enhancement on the target accident vehicle image, and generate a target enhancement image.

Referring to fig. 7, in a second embodiment of the image enhancement apparatus according to the present invention, the image enhancement apparatus specifically includes:

In this embodiment, the image enhancement device further includes:

a second obtaining module 605, configured to obtain a historical vehicle image of the accident scene from a preset database;

a construction module 606 for constructing a training sample dataset using the historical vehicle images as training samples;

The training module 607 is configured to input the training sample data set into a preset image enhancement generator, and train the image enhancement generator to obtain a target image enhancement model.

In this embodiment, the image enhancement device further includes:

the input module 608 is configured to obtain a plurality of training pictures from a preset training picture database, and preset a neural network model for the training pictures to obtain a prediction type label of the training pictures;

the generating module 609 is configured to generate a real category label of the training picture according to the category of the training picture; generating a first loss function of the training picture according to the prediction type label of the training picture and the real type label of the training picture;

and the updating module 610 is configured to update parameters in the neural network model according to the first loss function, so as to obtain a target picture identification model.

In this embodiment, the training module 607 is specifically configured to:

the image enhancement unit 6071 is configured to perform image enhancement on the training sample data set to obtain a preset number of target original images and corresponding enhanced images;

an input unit 6072 for inputting the target original image into a CNN image enhancement network in a preset image enhancement generator to obtain an output image corresponding to the target original image;

A determining unit 6073 for determining a second loss function between the output image and the enhanced image;

and the iterative training unit 6074 is configured to perform, based on the second loss function, on the CNN image enhancement network in the image enhancement generator until the CNN image enhancement network converges to obtain a target image enhancement model.

In this embodiment, the image enhancement module 605 is specifically configured to:

In this embodiment, the image enhancement device further includes:

a third obtaining module 611, configured to obtain owner identity information of a user and damage information of the accident vehicle, where the vehicle information includes vehicle risk information of the accident vehicle;

The prediction module 612 is configured to receive the claim settlement request uploaded by the user, and input the target enhanced image into the trained convolutional neural network model for prediction, so as to obtain the claim settlement probability of the target enhanced image;

a second judging module 613, configured to judge whether the claim probability of the target enhanced image is greater than a preset threshold;

a determining module 614, configured to determine that the target enhanced image is claimable when the claimable probability of the target enhanced image is greater than a preset threshold;

and a claim settlement module 615, configured to settle a claim for the accident vehicle based on the owner identity information of the user and the identification information of the accident vehicle.

The image enhancement apparatus in the embodiment of the present invention is described in detail above in fig. 6 and 7 from the point of view of the modularized functional entity, and the image enhancement device in the embodiment of the present invention is described in detail below from the point of view of hardware processing.

Fig. 8 is a schematic structural diagram of an image enhancement device according to an embodiment of the present invention, where the image enhancement device 800 may have a relatively large difference due to different configurations or performances, and may include one or more processors (central processing units, CPU) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) storing application programs 833 or data 832. Wherein memory 820 and storage medium 830 can be transitory or persistent. The program stored in the storage medium 830 may include one or more modules (not shown), each of which may include a series of instruction operations in the image enhancement device 800. Still further, the processor 810 may be configured to communicate with the storage medium 830 and execute a series of instruction operations in the storage medium 830 on the image enhancement device 800 to implement the steps of the image enhancement method provided by the method embodiments described above.

The image enhancement device 800 may also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input/output interfaces 860, and/or one or more operating systems 831, such as Windows Serve, mac OS X, unix, linux, freeBSD, and the like. It will be appreciated by those skilled in the art that the image enhancement device structure shown in fig. 8 is not limiting of the image enhancement device provided by the present application and may include more or fewer components than shown, or may combine certain components, or may be a different arrangement of components.

The present application also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, or may be a volatile computer readable storage medium, having stored therein instructions that, when executed on a computer, cause the computer to perform the steps of the above-described image enhancement method.

The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An image enhancement method, the image enhancement method comprising:

acquiring scene image data of a traffic accident scene;

if not, prompting the user to shoot again;

inputting the target accident vehicle image into a preset image enhancement model, and enhancing the target accident vehicle image to generate a target enhancement image;

acquiring owner identity information of a user and damage information of the accident vehicle; receiving a claim settlement request uploaded by the user, and inputting the target enhanced image into a trained convolutional neural network model for prediction to obtain the claim settlement probability of the target enhanced image; judging whether the claim settling probability of the target enhanced image is larger than a preset threshold value; if the claim settlement probability of the target enhanced image is larger than a preset threshold value, determining that the target enhanced image can settle the claim; based on the owner identity information of the user and the identification information of the accident vehicle, the accident vehicle is subjected to claim settlement;

Inputting the target accident vehicle image into a preset image enhancement model, enhancing the target accident vehicle image, and generating a target enhancement image comprises the following steps:

inputting the downsampled image into a target image enhancement model to obtain image enhancement data corresponding to the target accident vehicle image;

2. The image enhancement method according to claim 1, further comprising, prior to said acquiring the scene image data of the traffic accident scene:

3. The image enhancement method according to claim 1, further comprising, before said recognizing the live image data by a preset image recognition model, judging whether the live image data satisfies a preset image standard:

and updating parameters in the neural network model according to the first loss function to obtain a target image recognition model.

4. The image enhancement method according to claim 2, wherein inputting the training sample data set into a preset image enhancement generator, training the image enhancement generator, and obtaining a target image enhancement model comprises:

5. An image enhancement device, characterized in that the image enhancement device comprises:

The image enhancement module is used for inputting the target accident vehicle image into a preset image enhancement model, and carrying out image enhancement on the target accident vehicle image to generate a target enhancement image;

the prediction module is used for acquiring the owner identity information of the user and the damage information of the accident vehicle; receiving a claim settlement request uploaded by the user, and inputting the target enhanced image into a trained convolutional neural network model for prediction to obtain the claim settlement probability of the target enhanced image; judging whether the claim settling probability of the target enhanced image is larger than a preset threshold value; if the claim settlement probability of the target enhanced image is larger than a preset threshold value, determining that the target enhanced image can settle the claim; based on the owner identity information of the user and the identification information of the accident vehicle, the accident vehicle is subjected to claim settlement;

the image enhancement module is specifically configured to:

6. The image enhancement device of claim 5, further comprising:

7. The image enhancement device of claim 5, further comprising:

the input module is used for acquiring a plurality of training pictures from a preset training picture database, inputting the training pictures into a preset neural network model and obtaining a prediction type label of the training pictures;

and the updating module is used for updating parameters in the neural network model according to the first loss function to obtain a target image recognition model.

8. The image enhancement device of claim 6, wherein the training module comprises:

9. An electronic device, the electronic device comprising: a memory and at least one processor, the memory having instructions stored therein, the memory and the at least one processor being interconnected by a line;

the at least one processor invoking the instructions in the memory to cause the electronic device to perform the steps of the image enhancement method of any of claims 1-4.

10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the image enhancement method according to any of claims 1-4.