CN112202919A

CN112202919A - Picture ciphertext storage and retrieval method and system under cloud storage environment

Info

Publication number: CN112202919A
Application number: CN202011138837.4A
Authority: CN
Inventors: 陈驰; 原方圆; 田雪; 王佳宁; 苏帅
Original assignee: Institute of Information Engineering of CAS
Current assignee: Institute of Information Engineering of CAS
Priority date: 2020-10-22
Filing date: 2020-10-22
Publication date: 2021-01-08
Anticipated expiration: 2040-10-22
Also published as: CN112202919B

Abstract

The invention provides a method and a system for storing and retrieving a picture ciphertext in a cloud storage environment, wherein the method comprises a user side, a server side and a cloud server side: the user side is used for uploading the pictures or the query keywords to the server side; the server side is used for acquiring a feature tag set of the picture, encrypting the picture and the feature tag set and storing the obtained ciphertext picture and the ciphertext feature tag set to the cloud server side; sending the generated ciphertext query request to a cloud server according to the query keyword; decrypting the ciphertext picture sent by the cloud server side, and returning the obtained plaintext picture to the user side; the cloud server is used for storing the ciphertext picture and the ciphertext feature tag set and retrieving the ciphertext picture through the ciphertext query request. The method and the device protect private information in the picture from being leaked, realize the function of picture retrieval through the text keywords after the picture is encrypted and uploaded, and improve the retrieval efficiency of the image in a ciphertext environment through the retrieval feature tag.

Description

Picture ciphertext storage and retrieval method and system under cloud storage environment

Technical Field

The invention relates to the technical field of picture processing and the field of information security, in particular to a picture ciphertext storage and retrieval method and a picture ciphertext storage and retrieval system in a cloud storage environment.

Background

With the development of more than ten years, cloud computing has gradually fallen to applications from concepts, and various cloud services are provided for more and more enterprises. Cloud storage is a data storage infrastructure of cloud computing, and provides data storage services of acquisition on demand, elastic expansion, high cost performance and convenient access for government departments, scientific research institutions, enterprises in various rows and individual users. While the cloud storage is widely applied, the problem of cloud storage security is increasingly highlighted. For example, a large amount of image data in which relevant privacy information of the user may be contained is stored in the cloud storage. Currently, cloud storage service providers often adopt an encryption method to ensure that user privacy data is not leaked. However, as the storage scale of cloud images increases, how to quickly and efficiently retrieve encrypted images becomes a problem which needs to be solved urgently.

Published patent application CN103744976A provides an image security retrieval method based on Paillier homomorphic encryption. Because the homomorphic encryption technology can directly calculate and compare the encrypted data, the homomorphic encryption technology is used for encrypting the image characteristics when the image is uploaded, and the encrypted image characteristics can be retrieved without decryption. The implementation process of the patent is as follows: firstly, extracting relevant features from an image in cloud storage, and performing dimension reduction processing on the image features by using an LPP (low power point) method; then, carrying out encryption calculation on the image characteristics by adopting a Paillier homomorphic encryption algorithm; and finally, performing similarity matching on the picture to be searched and the encrypted image characteristics, and returning the most similar K images as a retrieval result. However, the dimension reduction of the image features can lose the content information of the image and reduce the accuracy of the retrieval result. Meanwhile, the method searches similar pictures in a picture searching mode, limits the application range to a certain extent and has low practical value.

The published patent application CN107480163A provides an efficient ciphertext image retrieval method supporting privacy protection in a cloud environment, which uses a layered K-means algorithm to establish an index based on a feature vector of an image, encrypts and uploads a picture and a picture index to a cloud storage, and retrieves a ciphertext picture through the picture index. The method improves the retrieval efficiency, but in the retrieval stage, the first k pictures with similar characteristics are retrieved in a picture searching mode, no systematic operation flow is provided, and the practical value is low.

Disclosure of Invention

Aiming at the technical problems in the prior art, the invention discloses a picture ciphertext storage and retrieval method and a picture ciphertext storage and retrieval system in a cloud storage environment.

The technical scheme of the invention is as follows:

a picture ciphertext storage method under a cloud storage environment is suitable for a network system consisting of a cloud server, a plurality of servers and a plurality of clients, and comprises the following steps:

1) receiving a picture of a user side, and acquiring a feature tag set of the picture;

2) encrypting the picture and the feature tag set by using the user key of the user side to generate a ciphertext picture and a ciphertext feature tag set;

3) and storing the ciphertext picture and the ciphertext feature tag set to a cloud server side.

Further, acquiring a feature tag set of the picture by the following steps:

1) converting the picture into a TFRecord form;

2) inputting the converted picture into one or more feature tag extraction models, and carrying out target detection to obtain a feature tag set of the picture;

the feature tag extraction model is obtained through the following steps:

a) collecting sample pictures, converting each sample picture into a TFrecord form, and labeling each sample picture;

b) classifying the converted sample pictures according to the labels;

b) inputting the converted sample pictures of the same category into the same convolutional neural network and learning to obtain a feature tag extraction model.

Further, the method of target detection comprises: and (3) a single-item multi-frame target detection algorithm.

Further, the method for collecting the sample picture comprises the following steps: using a web crawler method and collecting public data sets.

Further, the convolutional neural network is constructed by a Tensorflow machine learning framework.

Further, the architecture of the convolutional neural network comprises: and removing the image classification network and the multi-scale feature detection network of the classification layer.

Further, generating a ciphertext picture and a ciphertext feature tag set of the picture by the following steps:

1) performing data encapsulation on the picture and ciphertext feature tag set to obtain encapsulated data;

2) acquiring an encryption and decryption key and an encryption and decryption algorithm of a corresponding user side;

3) and generating a ciphertext picture and a ciphertext feature tag set of the picture by packaging the data, the encryption and decryption key and an encryption and decryption algorithm.

Further, the encryption and decryption algorithm comprises: opennssl algorithm.

A picture ciphertext retrieval method in a cloud storage environment is a network system consisting of a cloud server, a plurality of servers and a plurality of clients, and comprises the following steps:

1) receiving a query keyword of a client, and encrypting the query keyword through a user key of the client to obtain a keyword ciphertext;

2) generating a ciphertext query request according to the keyword ciphertext;

3) receiving a ciphertext picture sent by a cloud server, wherein the ciphertext picture is obtained by retrieving a ciphertext feature tag set stored in the cloud server by using a ciphertext query request;

4) and decrypting the ciphertext picture by using the user key of the user side, and returning the obtained plaintext picture to the user side.

A picture ciphertext retrieval system in a cloud storage environment, comprising:

the user side is used for uploading the pictures or the query keywords to the server side; receiving a plaintext picture sent by a server;

the server is used for receiving the picture of the user side and acquiring a feature tag set of the picture; encrypting the picture and the feature tag set by using a user key of a user side to generate a ciphertext picture of the picture and a ciphertext feature tag set; storing the ciphertext picture and the ciphertext feature tag set to a cloud server side; receiving a query keyword of a client, and encrypting the query keyword through a user key of the client to obtain a keyword ciphertext; generating a ciphertext query request according to the keyword ciphertext; receiving a ciphertext picture sent by a cloud server; decrypting the ciphertext picture by using a user key of the user side, and returning the obtained plaintext picture to the user side;

the cloud server is used for storing the ciphertext picture and the ciphertext feature tag set; and searching the ciphertext feature tag set stored at the cloud server by using the ciphertext query request to obtain a ciphertext picture.

Compared with the prior art, the method and the system for searching the picture ciphertext under the cloud storage environment have the following advantages:

1. according to the method and the system for retrieving the picture ciphertext in the cloud storage environment, the picture is encrypted and then uploaded to the cloud server, and privacy information in the picture is protected from being leaked. Meanwhile, the function of picture retrieval through the text keywords is realized after the picture is encrypted and uploaded, and the problem of picture content retrieval in a ciphertext environment is solved.

2. According to the method and the system for retrieving the picture ciphertext under the cloud storage environment, the feature tag of the picture is generated by using target detection in the picture uploading process, and the picture is retrieved by retrieving the feature tag in the picture retrieval stage, so that the retrieval efficiency of the picture under the ciphertext environment is improved, and quick and efficient retrieval is realized.

Drawings

Fig. 1 is a flowchart illustrating an embodiment of a method and a system for retrieving a picture ciphertext in a cloud storage environment.

Fig. 2 is a structural diagram of a method and a system for retrieving a picture ciphertext in a cloud storage environment according to the present invention.

FIG. 3 is a training flow diagram of a picture ciphertext retrieval method and a system feature model in a cloud storage environment.

FIG. 4 is a SSD model diagram for image ciphertext retrieval in a cloud storage environment, according to the method and system for target detection.

FIG. 5 is a picture encryption flowchart of a method and system for retrieving a picture ciphertext in a cloud storage environment, in accordance with the present invention.

FIG. 6 is a picture decryption flowchart of a method and system for retrieving a picture ciphertext in a cloud storage environment, according to the present invention.

FIG. 7 is a picture retrieval flow chart of a method and system for retrieving a picture ciphertext in a cloud storage environment, according to the present invention.

FIG. 8 is a comparison line graph of the number of pictures and the retrieval time in the case of retrieving the plaintext according to the method and system for retrieving the picture ciphertext in the cloud storage environment.

Detailed Description

The present invention is further illustrated in detail below with reference to specific examples, but the scope of the present invention is not limited in any way.

As shown in fig. 1, the system mainly includes a user side, a server side and a cloud server side, the user side uploads pictures and inputs feature characters to perform picture retrieval operation, the server side needs to train a model of a target object in advance, when the user uploads the pictures, the trained model is used to perform target detection on the pictures, generate feature tags of the pictures and encrypt the pictures and the feature tags of the pictures, and the encrypted information is uploaded to the cloud server side; when a user inputs characteristic characters to perform picture retrieval, the server side retrieves the characteristic labels in the cloud server side to obtain a retrieval result, decrypts the retrieved result and returns the decrypted result to the user side, and the functions of picture encryption storage and encryption retrieval are achieved. And the cloud server side is responsible for storing the encrypted picture file and the feature tag of the picture.

When a user uploads a picture, the method comprises the following five steps:

the method comprises the following steps: performing model training on the collected training set images by using a Tensorflow machine learning framework to obtain a relevant feature model Mi … Mj, such as feature models of airplanes, automobiles and the like;

step two: a user logs in the system and executes the related operation of uploading the image Pi;

step three: the server side performs target detection on the image Pi by using a characteristic model Mi., Mj through an SSD target detection algorithm, and generates a characteristic label Li … Lj of the image Pi;

step four: the picture Pi and the feature tag Li … Lj thereof are encrypted using the user key, and ciphertext information CPi and a ciphertext feature tag CLi … CLj of the picture are generated.

Step five: and uploading the ciphertext information CPi and the ciphertext feature tag CLi … CLj of the picture to a cloud server side.

When a user retrieves a picture, the method comprises the following five steps:

the method comprises the following steps: and (4) logging in the system by a user, and inputting the characteristic characters W to perform picture retrieval, such as retrieving pictures related to airplanes.

Step two: and the server side encrypts the characteristic character W by using the user key to generate a ciphertext characteristic CW.

Step three: searching in ciphertext feature tag of picture stored in cloud server using ciphertext feature CW

Step four: the cloud server returns the retrieved picture CPi to the server side

Step five: and the server side decrypts the CPi by using the user key and returns the decrypted plaintext image Pi to the user side. The system can realize picture ciphertext retrieval in the cloud storage environment.

From the steps, the defects and risks of the prior art scheme mentioned in the background technology are solved perfectly through the method and the system for retrieving the picture ciphertext under the cloud storage environment, the pictures stored to the cloud of the third party are all in the form of the ciphertext, a cloud service provider cannot maliciously snoop data of a user, and the data are not easy to eavesdrop in the process of uploading the pictures.

The system mainly comprises a training model module, a picture target detection and feature tag generation module, a picture encryption and decryption module and a picture retrieval module, as shown in figure 2.

The specific functions of each module are as follows:

and training the model module. Training models of target objects such as vehicles, airplanes and the like, wherein the trained models are used in a target detection and labeling stage.

And the image target detection and generation characteristic label module. And carrying out target detection on the image uploaded by the user to generate a feature tag of the picture.

And the picture encryption and decryption module. The picture and the feature tag uploaded to the cloud server by the user are encrypted and uploaded to the cloud server, so that the information of the picture is prevented from being leaked; and decrypting and opening the ciphertext picture retrieved by the user to obtain the plaintext picture.

And a picture retrieval module. And retrieving the image feature tag in the cloud server, and returning the retrieved picture to the user.

Further, each module of the method and the system for retrieving the picture ciphertext in the cloud storage environment is specifically analyzed.

In the training model module, model training is performed by using a Tensorflow machine learning framework, as shown in FIG. 3, which comprises the following five steps:

step 1: a training set is obtained. And acquiring pictures through the web crawler and the public data set, and labeling the target objects in the training set.

Step 2: and (5) performing preliminary processing on the image. TensorFlow provides a uniform format for storing data, i.e., TFRecord. It is therefore necessary to convert the pictures in the training set to TFRecord form for the purpose of subsequent model training.

And step 3: and building a convolutional neural network model. And (3) constructing an initial convolutional neural network aiming at the input information and the picture label information through the analysis of the image.

And 4, step 4: and (5) training a model. And taking the training image as input, combining the label of the training image, training and optimizing the constructed initial convolutional neural network to obtain a target model, and storing the training parameters for the next use.

And 5: and (5) testing the model. The trained model is tested using a test set to check the accuracy of the model.

In the image target detection and feature tag generation module, a core algorithm for identifying a target object in an image is a Single Shot multi box Detector (SSD) target detection algorithm. As shown in fig. 4, the SSD architecture is mainly divided into two parts, one part is a deep convolutional neural network located at the front end, and an image classification network with a classification layer removed, such as VGG, is used for the preliminary feature extraction of the target; the other part is a multi-scale feature detection network positioned at the rear end, which is a group of cascaded convolutional neural networks, and feature extraction is carried out on a feature layer generated by the front-end network under the condition of different scales, so that the targets of detecting different sizes are achieved. For one picture, the SSD outputs a detection box of the destination, and a category of the destination. And the SSD adopts a multi-scale feature map for detection, a large-scale feature map can be divided into more small units, and the prior frame of each unit is smaller and is used for detecting small targets. The small-scale feature map can be divided into larger units, and the scale of the prior frame of each unit is larger and used for detecting a large target. The SSD sets prior frames with different length-width ratios on each unit, the prior frames of the units are used for predicting the bounding box, and the prior frame matched with the shape of the actual target object can be found during training. According to the SSD scheme, a picture is input, the position of a picture target, namely a bounding box, is directly regressed, and the target is classified to generate a feature tag. The method is based on SSD target detection, and the feature labels of the uploaded pictures are obtained by using the object model trained by the training model module and the SSD target detection algorithm.

As shown in fig. 5, the image encryption and decryption module includes four modules, namely, an image package module, an encryption and decryption configuration module, a key acquisition module, and a specific encryption and decryption module. The specific encryption process is as follows:

step 1, transmitting the plaintext picture and the feature tag of the picture into a data encapsulation module through a picture encryption and decryption module.

And 2, partitioning and packaging the transmitted picture and the feature tag thereof to obtain a uniform format.

And 3, requesting to query the encryption and decryption algorithm stored by the user from the encryption and decryption configuration module through the picture encryption and decryption module.

And 4, the encryption and decryption configuration module accesses the cached configuration file and returns the inquired encryption and decryption algorithm to the picture encryption and decryption module.

And 5, requesting to acquire an encryption and decryption key from a user through the picture encryption and decryption module.

And 6, the user accesses the encryption and decryption keys stored in the memory cache and returns the encryption and decryption keys to the picture encryption and decryption module.

And 7, calling a concrete implementation interface of image encryption and decryption, and transmitting a plaintext, an algorithm type and an encryption and decryption key as parameters.

And 8, specifically, the encryption and decryption can be realized by an open source algorithm library such as openssl and the like or a user-defined function, and the block plaintext is encrypted and returned.

The picture decryption process is the same as the above encryption process, as shown in fig. 6. The specific decryption steps are as follows:

step 1, requesting to inquire an encryption and decryption algorithm stored by a user from an encryption and decryption configuration module through a picture encryption and decryption module.

And 2, the encryption and decryption configuration module accesses the cached configuration file and returns the inquired encryption and decryption algorithm to the picture encryption and decryption module.

And 4, requesting to acquire the encryption and decryption key from the key management module through the picture encryption and decryption module.

And 5, the key management module accesses the encryption and decryption keys stored in the memory cache and returns the encryption and decryption keys to the picture encryption and decryption module.

And 6, calling a concrete implementation interface of image encryption and decryption, and transmitting the ciphertext image, the algorithm type and the encryption and decryption key as parameters.

And 7, the specific encryption and decryption can be realized by an open source algorithm library such as openssl and the like or a user-defined function, and the ciphertext is decrypted and returned.

And 8, transmitting the decrypted packaged picture into a data packaging module through the picture encryption and decryption module.

And 9, converting the decrypted packaged picture into the original picture by the data packaging module and returning the original picture to the picture encryption and decryption module.

And the picture retrieval module is used for retrieving the image feature tags in the cloud server and returning the retrieved pictures to the user. As shown in fig. 7, the specific steps are as follows:

step 1, a user inputs picture keywords to be searched in a system query interface, and a system defaults to ciphertext query.

And 2, encrypting the query keyword by using a user key, namely calling an encryption algorithm to obtain a ciphertext of the keyword.

And 3, generating a ciphertext query statement according to the ciphertext of the keyword, and sending a ciphertext query request in a fixed format to the cloud server by the server side.

And 4, after the cloud server calls and receives the query request, performing query processing to obtain the queried ciphertext picture.

And 5, the cloud server returns the inquired ciphertext picture to the server side, and the server side calls a picture encryption and decryption model to decrypt and open the picture on an inquiry result interface.

An example of the application of the present invention is given below.

1000 relevant images of the airplane are collected from a Corel image library to serve as a training set, an airplane model is generated by adopting the method, a plurality of images are collected from a hundred-degree image searching website and are encrypted and uploaded to a cloud server, a user searches the relevant images of the airplane by adopting the method, and the relation between the searching time and the number of the images is shown in figure 8 by comparing the searching time and the retrieving time of the images in the plaintext. The image retrieval time of the image ciphertext storage and retrieval method provided by the invention is related to the number of the images, and meanwhile, the difference between the encrypted image retrieval time and the plaintext image retrieval time is not large, so that the quick retrieval of the ciphertext image can be realized.

The above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and a person skilled in the art can make modifications or equivalent substitutions to the technical solution of the present invention without departing from the spirit and scope of the present invention, and the scope of the present invention should be determined by the claims.

Claims

1. A picture ciphertext storage method under a cloud storage environment is suitable for a network system consisting of a cloud server, a plurality of servers and a plurality of clients, and comprises the following steps:

2. The method of claim 1, wherein the feature tag set of the picture is obtained by:

1) converting the picture into a TFRecord form;

the feature tag extraction model is obtained through the following steps:

b) classifying the converted sample pictures according to the labels;

3. The method of claim 2, wherein the method of target detection comprises: and (3) a single-item multi-frame target detection algorithm.

4. The method of claim 2, wherein collecting the sample picture comprises: using a web crawler method and collecting public data sets.

5. The method of claim 2, wherein the convolutional neural network is constructed by a Tensorflow machine learning framework.

6. The method of claim 2, wherein the architecture of the convolutional neural network comprises: and removing the image classification network and the multi-scale feature detection network of the classification layer.

7. The method of claim 1, wherein the ciphertext picture and the ciphertext feature tag set of the picture are generated by:

8. The method of claim 7, wherein the encryption and decryption algorithm comprises: opennssl algorithm.

9. A picture ciphertext retrieval method under a cloud storage environment is suitable for a network system consisting of a cloud server, a plurality of servers and a plurality of clients, and comprises the following steps:

2) generating a ciphertext query request according to the keyword ciphertext;

3) receiving a ciphertext picture sent by a cloud server, wherein the ciphertext picture is obtained by retrieving a ciphertext feature tag set stored in the cloud server by any one of the methods of claims 1-8 by using a ciphertext query request;

10. A picture ciphertext retrieval system in a cloud storage environment, comprising: