CN113660386A - Color image encryption compression and super-resolution reconstruction system and method - Google Patents
Color image encryption compression and super-resolution reconstruction system and method Download PDFInfo
- Publication number
- CN113660386A CN113660386A CN202110781008.6A CN202110781008A CN113660386A CN 113660386 A CN113660386 A CN 113660386A CN 202110781008 A CN202110781008 A CN 202110781008A CN 113660386 A CN113660386 A CN 113660386A
- Authority
- CN
- China
- Prior art keywords
- image
- coding sequence
- encryption
- module
- reconstruction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
- H04N1/32272—Encryption or ciphering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/44—Secrecy systems
Abstract
The invention discloses a color image encryption compression and super-resolution reconstruction system and method, wherein the system comprises a preprocessing module, an encryption compression module, a decryption module and a reconstruction module; the method comprises the steps that an original image is input into a preprocessing module, the preprocessing module carries out preprocessing operation on the original image to obtain a coding sequence, and then the original image and the coding sequence are input into an encryption compression module; the encryption compression module encrypts and compresses the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence, and sends the encrypted compressed image file and the encrypted coding sequence to the decryption module; the decryption module carries out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image, and sends the interpolation image and the decrypted image to the reconstruction module; the reconstruction module acquires local features, a feature weight graph and an up-sampling image of the interpolation image, fuses to generate an initial reconstruction image, corrects the initial reconstruction image by using the decryption image, and outputs a final reconstruction image. The invention can effectively improve the reconstruction quality of the reconstructed image.
Description
Technical Field
The present invention relates to the field of image processing, and more particularly, to a system and method for color image encryption compression and super-resolution reconstruction.
Background
In the conventional communication, in order to reduce the data transmission and storage cost, image and video data are compressed and encrypted to ensure safety, and are finally transmitted to a receiving party through a public channel. The receiver receives the compressed and encrypted data, and then performs decryption and decompression, and a decryption key is obtained from a secure channel, as shown in fig. 1, and the process is called a compression before encryption (CTE) system; with the development of cloud computing, a large number of users and enterprises may choose to store data in the cloud in order to reduce data storage and management costs. In order to solve the security problem caused by data storage in a cloud end, encryption is carried out before data is uploaded, and compression is carried out after the data is uploaded to the cloud end. After a user obtains a key through a secure channel, decrypting the encrypted and then compressed ciphertext by using the key, and then reconstructing the decrypted ciphertext; as shown in fig. 2, this process of encryption before compression is referred to as an ETC system. However, the encryption operation masks the data content and thus the statistical properties of the data to be compressed, so that it is intuitively difficult to encrypt the data for compression. Johnson et al demonstrate through information demonstration that an ETC system theoretically has the same compression and safety performance as a CTE system under lossless compression, and approaches the compression performance of the CTE system under lossy compression. Many researchers have conducted intensive research because lossy compression of image and video data can be balanced between a lower compression rate and higher visual quality. Compression algorithms for lossy compression of encrypted data can be classified into three categories, namely, encryption lossy compression methods based on Compressive Sensing (CS), scalar quantization (scalar quantization), and uniform downsampling. Among the compression methods, uniform down sampling is a simple and efficient compression technology, and is very suitable for compression processing at a cloud end where a decryption key cannot be obtained. However, this compression method has a problem of losing a large amount of image data compared with other methods, and eventually, the reconstruction quality is difficult to achieve a satisfactory level.
Since the decrypted uniformly downsampled data can be regarded as a Low Resolution (LR) image of the original image, the problem that the encrypted image is compressed by uniformly downsampling and the receiver recovers the original image by using the decrypted data can be equivalent to a Super Resolution (SR) problem. Many methods have been proposed to solve the problem of compressing encrypted images, but most methods only aim at grayscale images, and few methods aim at color images; and SR models based on deep learning have some drawbacks: for example, high-frequency details of an image are not well recovered, different spaces and channels are viewed identically, and feature information is not distinguished and fully utilized, so that the improvement of SR performance is limited.
Chinese patent CN110084745A published in 8/2/2019 provides a parallel single-frame image super-resolution reconstruction method based on a dense convolutional neural network, which includes: and constructing a dense convolutional neural network consisting of two parallel dense connection structure blocks and a jump layer connection structure, wherein the two parallel dense connection structure blocks are used as cyclic sub-blocks, and each branch comprises a sub-block structure and identity mapping. The data base reconstructed by the method only comprises the characteristic information of the low-resolution picture, and the reconstruction quality is reduced because the peripheral information of the pixel points is not concerned; during reconstruction, shallow features are added and connected with deep features through a jump layer connection structure, so that part of learning information is lost; and for local features of different spaces and channels, the high-frequency feature information is not distinguished and fully utilized, so that key details of a reconstructed image cannot be completely recovered, and the reconstruction quality is poor.
Disclosure of Invention
The invention provides a color image encryption compression and super-resolution reconstruction system and method for overcoming the defect of poor super-resolution reconstruction quality of an encrypted color image in the prior art, and the high-frequency characteristic information is fully utilized when the super-resolution reconstruction is carried out on the encrypted color image, so that the pixel of the reconstructed image is closer to the pixel of the original image, and the reconstruction quality of the reconstructed image is effectively improved.
In order to solve the technical problems, the technical scheme of the invention is as follows:
the invention provides a color image encryption compression and super-resolution reconstruction system, which comprises a preprocessing module, an encryption compression module, a decryption module and a reconstruction module;
the method comprises the steps that an original image is input into a preprocessing module, the preprocessing module carries out preprocessing operation on the original image to obtain a coding sequence, and then the original image and the coding sequence are input into an encryption compression module; the encryption compression module encrypts and compresses the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence, and sends the encrypted compressed image file and the encrypted coding sequence to the decryption module; the decryption module carries out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image, and sends the interpolation image and the decrypted image to the reconstruction module; the reconstruction module acquires local features, a feature weight graph and an up-sampling image of the interpolation image, fuses to generate an initial reconstruction image, corrects the initial reconstruction image by using the decryption image, and outputs a final reconstruction image.
Preferably, the preprocessing module performs a preprocessing operation on the original image specifically as follows:
the preprocessing module is used for input original image IorPerforming interpolation operation and downsampling operation to obtain interpolated image IbicAnd downsampling of subgraph Iordwn；
Pixel-by-pixel computed interpolation image IbicWith down-sampled subgraph IordwnError file E ofimgFor error file EimgCoding to obtain coding sequence Eac；
Coding sequence EacAttached to the original image IorAnd after the end, sending the data to an encryption compression module.
Preferably, the input original image I is interpolated using bicubic interpolationorAn interpolation operation is performed.
Preferably, the encryption compression module comprises an encryption unit and a compression unit;
the encryption unit uses the encryption key to encrypt the received original image IorAnd coding sequence EacThe encryption operation is carried out and the encryption operation is carried out,obtaining an encrypted image file IencAnd an encrypted coding sequence EencTo encrypt an image file IencAnd an encrypted coding sequence EencSending the key to the compression unit and sending the encryption key used for encryption to the decryption module;
the compression unit encrypts the image file IencCompressing to obtain a compressed image file Idwn(ii) a To compress an image file IdwnAnd an encrypted coding sequence EencAnd sending the data to a decryption module.
Preferably, the decryption module performs decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain the interpolated image and the decrypted image by the specific method comprising:
the decryption module is used for decrypting the compressed image file I according to the received encryption keydwnAnd an encrypted coding sequence EencPerforming decryption operation to obtain decrypted image IdecAnd coding sequence Eac；
For coding sequence EacDecoding to obtain an error file Eimg；
Will decrypt the image IdecAnd an error file EimgAdding pixel by pixel to obtain an interpolation image Ibic。
Preferably, the reconstruction module comprises a residual error intensive network unit, a global jump connection unit, a spatial attention network unit and a modification construction unit;
the space attention network unit endows different weights to the features of the interpolation image at different space positions to obtain a feature weight graph, and the feature weight graph is output to the residual error dense network unit;
the global jump connection unit is used for up-sampling the interpolation image to obtain an up-sampled image, and jump-connecting the up-sampled image to the residual error dense network unit;
the residual dense network unit is used for extracting local features of the interpolation image and fusing the local features and the feature weight graph into local weight fusion features; adding the local weight fusion features and the up-sampled image to generate an initial reconstruction image, and outputting the initial reconstruction image to a correction construction unit;
the correction construction unit corrects the initial reconstructed image by using the decrypted image to obtain a final reconstructed image.
Preferably, the spatial attention network unit is improved based on a traditional U-Net network, and comprises: the BN layer in the conventional U-Net network is removed and the deconvolution replaced with a sub-pixel convolution.
The conventional U-Net network adopts a BN (batch normalization) layer, which is beneficial to maintaining the stable distribution of gradients in the network and accelerating the convergence speed of a model, but also changes the statistical correlation among characteristic information, thereby causing the performance reduction of super-resolution; secondly, the traditional U-Net network adopts deconvolution to perform up-sampling coding, so that the images are uneven and the quality of the images obtained by super-resolution is also influenced; in order to keep the statistical correlation between the features and the pixels, the BN layer is deleted; in order to obtain better high-frequency details, deconvolution is replaced by sub-pixel convolution, and the reconstruction quality of the super-resolution reconstruction image is greatly improved.
Preferably, the residual dense network unit includes a first convolutional layer, a second convolutional layer, D residual dense blocks, a fusion layer, a 1 × 1 convolutional layer, a third convolutional layer, a fusion point, an upsampling layer, a summing point, and a fourth convolutional layer, which are connected in sequence;
extracting shallow layer characteristics from the input interpolation image through a first convolution layer and a second convolution layer; d residual error dense blocks respectively extract the features of each layer, and each residual error dense block sends one extracted local feature to the fusion layer for local feature fusion; after the number of channels is reduced by 1 multiplied by 1 convolution and local fusion characteristics are further extracted by a third convolution layer, fusing the fusion points with the characteristic weight graph extracted by the space attention network unit to obtain local weight fusion characteristics; after the up-sampling layer is amplified, the up-sampling image obtained by the global jump connection unit is added at an addition point, and an initial reconstructed image is output through a fourth convolution layer.
The local features extracted from the previous residual error dense block can be directly connected with the current residual error dense block, so that the communication of information flow is ensured, and the local features extracted from each residual error dense block can be fused by a fusion layer and fully utilized; the spatial attention network unit endows the interpolation image with different weights to the features of different spatial positions, and fuses the feature weight image and the local fusion features, so that the local fusion features are distinguished in importance, and high-frequency information is fully utilized; the global jump connection unit directly performs up-sampling on the interpolation image, avoids loss of partial data, and performs addition operation on the up-sampled image and the partial weight fusion feature at an addition point, so that the reconstruction quality of the initial reconstruction image is more excellent.
Preferably, the number of convolution kernels of the first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer is 3 × 3.
Preferably, the error file E is encoded using arithmetic codingimgCoding to obtain coding sequence Eac(ii) a Coding sequence E using arithmetic codingacDecoding to obtain an error file Eimg。
The arithmetic coding has the advantages of high coding efficiency, good real-time performance, high flexibility and strong adaptability.
Preferably, the encryption unit uses a stream cipher-based encryption method for the received original image IorAnd coding sequence EacCarrying out encryption operation; the compression unit encrypts the image file I using uniform downsamplingencCompression is performed.
The stream cipher has the advantages of high absolute security, simple realization and high encryption and decryption processing speed; the uniform downsampling is simple and efficient, and is suitable for compression processing under the condition that an encryption key cannot be obtained.
Preferably, the encryption key used for encryption is sent to the decryption module through a secure channel to compress the image file IdwnAnd an encrypted coding sequence EencAnd sending the data to a decryption module through a common channel.
The invention also provides a color image encryption compression and super-resolution reconstruction method, which comprises the following steps:
s1: preprocessing an original image to obtain a coding sequence;
s2: carrying out encryption and compression operations on the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence;
s3: carrying out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image;
s4: obtaining local features, feature weight graphs and up-sampling images of the interpolation images, fusing to generate an initial reconstruction image, correcting the initial reconstruction image by using the decryption image, and outputting a final reconstruction image.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
the method aims at taking a color image as an original image, obtaining an interpolation image and a decryption image of the original image through a preprocessing module, an encryption compression module and a decryption module, wherein pixel points in the interpolation image comprise peripheral information of corresponding pixel points in the original image, the decryption image comprises characteristic information of the original image, the interpolation image and the decryption image are jointly used as input of a reconstruction module, and the quality of the reconstruction image is improved on the basis of data; the reconstruction module gives different weights to the features of the interpolation image at different spatial positions to obtain a feature weight map; fusing the feature weight graph and the extracted local features, and distinguishing and fully utilizing more important high-frequency feature information; then, the initial reconstructed image is generated by adding the acquired up-sampling image, so that the loss of characteristic information is reduced; and finally, the initial reconstructed image is corrected by utilizing the decrypted image, so that the pixel points of the final reconstructed image and the original image are kept the same, the reconstructed pixel is closer to the original pixel, the reconstruction quality of the final reconstructed image is greatly improved, the image is more accurate, and the reconstruction effect is better.
Drawings
FIG. 1 is a block diagram of a prior art compression-before-encryption system;
FIG. 2 is a block diagram of a prior art encryption followed by compression system;
FIG. 3 is a block diagram of a color image encryption compression and super-resolution reconstruction system according to embodiment 1;
FIG. 4 is a block diagram of a spatial attention network unit according to embodiment 1;
FIG. 5 is a structural diagram of a conventional U-Net network described in embodiment 1;
FIG. 6 is a block diagram of a residual error intensive network unit described in example 1;
FIG. 7 is a schematic diagram of the connection between the reconstruction module units according to embodiment 1;
fig. 8 is a flowchart of a color image encryption compression and super-resolution reconstruction method according to embodiment 2.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
The embodiment provides a color image encryption compression and super-resolution reconstruction system, as shown in fig. 3, the system includes a preprocessing module, an encryption compression module, a decryption module and a reconstruction module;
the method comprises the steps that an original image is input into a preprocessing module, the preprocessing module carries out preprocessing operation on the original image to obtain a coding sequence, and the original image and the coding sequence are input into an encryption compression module; the encryption compression module encrypts and compresses the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence, and sends the encrypted compressed image file and the encrypted coding sequence to the decryption module; the decryption module carries out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image, and sends the interpolation image and the decrypted image to the reconstruction module; the reconstruction module acquires local features, a feature weight graph and an up-sampling image of the interpolation image, fuses to generate an initial reconstruction image, corrects the initial reconstruction image by using the decryption image, and outputs a final reconstruction image.
Different from the traditional super-resolution task, the reconstruction module generates a final reconstruction image according to the interpolation image and the decryption image; in the traditional super-resolution task, only an interpolation image is used as an LR image to reconstruct an original image, so that various problems in the background art exist; only the decrypted image is used as the LR image, the performance index PSNR of the super-resolution is obviously reduced, so that the interpolation image and the decrypted image are simultaneously obtained to be used as the input of the reconstruction module, and the reconstruction quality of the reconstructed image can be greatly improved.
The preprocessing module is used for input original image IorPerforming interpolation operation and downsampling operation to obtain interpolated image IbicAnd downsampling of subgraph Iordwn；
Pixel-by-pixel computed interpolation image IbicWith down-sampled subgraph IordwnError file E ofimg：
Eimg(xi,yi)＝Ibic(xb,yb)-Iordwn(xo,yo)
In the formula (x)i,yi)、(xb,yb)、(xo,yo) Respectively representing the coordinates of the pixel points at the same position in the interpolation image, the down-sampling subgraph and the error file;
for error file EimgCoding to obtain coding sequence Eac；
Coding sequence EacAttached to the original image IorAnd after the end, sending the data to an encryption compression module.
In the present embodiment, the interpolation operation is bicubic interpolation; error file E using arithmetic codingimgCoding to obtain coding sequence Eac。
The encryption compression module comprises an encryption unit and a compression unit;
the encryption unit uses the encryption key to encrypt the received original image IorAnd coding sequence EacCarrying out encryption operation to obtain an encrypted image file IencAnd an encrypted coding sequence EencTo encrypt an image file IencAnd an encrypted coding sequence EencSending to the compression unit, and sending the encryption key for encryption to the decryption module through the secure channelA block;
the compression unit encrypts the image file IencCompressing to obtain a compressed image file Idwn(ii) a To compress an image file IdwnAnd an encrypted coding sequence EencAnd sending the data to a decryption module through a common channel.
In the present embodiment, the received original image I is encrypted using a stream cipher-based encryption methodorAnd coding sequence EacCarrying out encryption operation; the compression unit encrypts the image file I using uniform downsamplingencCompression is performed.
The decryption module is used for decrypting the compressed image file I according to the received encryption keydwnAnd an encrypted coding sequence EencPerforming decryption operation to obtain decrypted image IdecAnd coding sequence Eac(ii) a For coding sequence EacDecoding to obtain an error file Eimg(ii) a Will decrypt the image IdecAnd an error file EimgAdding pixel by pixel to obtain an interpolation image Ibic。
Interpolated image IbicData amount of (2) and original image IorThe data volume of the interpolation image I is the same, the transmission and storage occupy larger storage space and communication bandwidth, and the interpolation image I is completely restored after decryption by calculating a transmission error filebicNot only ensuring the interpolation image IbicThe integrity of the image reconstruction method ensures a data base for reconstructing the image, and can reduce storage space and communication bandwidth and improve transmission efficiency.
The reconstruction module comprises a residual error intensive network unit, a global jump connection unit, a space attention network unit and a correction construction unit;
the space attention network unit endows different weights to the features of the interpolation image at different space positions to obtain a feature weight graph, and the feature weight graph is output to the residual error dense network unit;
the global jump connection unit is used for up-sampling the interpolation image to obtain an up-sampled image, and jump-connecting the up-sampled image to the residual error dense network unit;
the residual dense network unit is used for extracting local features of the interpolation image and fusing the local features and the feature weight graph into local weight fusion features; adding the local weight fusion features and the up-sampled image to generate an initial reconstruction image, and outputting the initial reconstruction image to a correction construction unit;
the correction construction unit corrects the initial reconstructed image by using the decrypted image to obtain a final reconstructed image.
As shown in fig. 4, the spatial attention network unit is improved based on a conventional U-Net network, and includes: the BN layer in the conventional U-Net network is removed and the deconvolution replaced with a sub-pixel convolution.
As shown in fig. 5, the conventional U-Net network is a symmetric structure with encoding and decoding, and the features extracted by the convolutional layer are first successively down-sampled and encoded to obtain smaller features; and then, carrying out cascade up-sampling decoding on the features to obtain weight graphs with different weights at different spatial positions. And by utilizing the weight of the U-Net network, different differences can be given at different spatial positions.
As shown in fig. 6, the residual error dense network unit includes a first convolutional layer, a second convolutional layer, D residual error dense blocks, a fusion layer, a 1 × 1 convolutional layer, a third convolutional layer, a fusion point, an upsampling layer, a summing point, and a fourth convolutional layer, which are connected in sequence;
as shown in fig. 7, the shallow feature of the input interpolation image is extracted by the first convolution layer and the second convolution layer; d residual error dense blocks respectively extract the features of each layer, and each residual error dense block sends one extracted local feature to the fusion layer for local feature fusion; after the number of channels is reduced by 1 multiplied by 1 convolution and local fusion characteristics are further extracted by a third convolution layer, fusing the fusion points with the characteristic weight graph extracted by the space attention network unit to obtain local weight fusion characteristics; after the up-sampling layer is amplified, the up-sampling image obtained by the global jump connection unit is added at an addition point, and an initial reconstructed image is output through a fourth convolution layer.
The local features extracted from the previous residual error dense block can be directly connected with the current residual error dense block, so that the communication of information flow is ensured, and the local features extracted from each residual error dense block can be fused by a fusion layer and fully utilized; the spatial attention network unit endows the interpolation image with different weights to the features of different spatial positions, and fuses the feature weight image and the local fusion features, so that the local fusion features are distinguished in importance, and high-frequency information is fully utilized; the global jump connection unit directly performs up-sampling on the interpolation image, avoids loss of partial data, and performs addition operation on the up-sampled image and the partial weight fusion feature at an addition point, so that the reconstruction quality of the initial reconstruction image is more excellent.
The convolution kernels of the first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer are all 3 multiplied by 3.
In practical operation, the original image with the resolution of H × W is recorded as IorThe encrypted image file obtained by encrypting the original image based on the stream cipher is recorded as IencFor encrypted image file IencThe compressed image file obtained by uniform down-sampling (compression) is recorded as IdwnFor compressed image file IdwnThe decrypted image obtained is marked as Idec；IdwnIs IencUniformly downsampling the resulting image, IdecIs also equivalent to IencThe uniformly down-sampled image, which likewise corresponds to the starting image IorUniformly downsampling the obtained image; so that the decrypted image I can be useddecViewed as the original image IorBy decrypting the image IdecReconstructing an original image IorIs essentially regarded as a super-resolution (SR) problem; the goal of the SR is to reconstruct the original image IorThe closest final reconstructed image I', therefore, the reconstruction problem of the color image encryption compression and super-resolution reconstruction system proposed in this embodiment is expressed as:
s.t.I′(s×x1,s×y1)＝Idec(x0,y0)
in the formula, PSNR (-) represents the function of calculating the peak signal-to-noise ratio, H represents the height of the original picture, and W represents the width of the original picture，
Representing a set of natural numbers; s.t. denotes a constraint, s denotes a down-sampling factor, x1,y1Representing the abscissa and ordinate, x, of a pixel point on the final reconstructed image0,y0Representing the abscissa and ordinate of the pixel point on the decrypted image; the meaning of the constraint condition is that in the SR process, the coordinates of the pixel points on the low-resolution image are equal to those of the pixel points on the original image. The constraint condition is realized by a reconstruction module, so that the pixel point coordinate on the final reconstructed image is kept the same as the pixel point coordinate on the original image, and the reconstructed pixel is closer to the original pixel.
Considering that the key of SR is to recover enough high-frequency information from LR image, regions such as edges, textures, etc. in the image should be given greater weight, and a spatial attention network unit is introduced based on this; the spatial attention network unit is improved based on a traditional U-Net network; however, the application of the traditional U-Net network to the SR task has the following disadvantages: the conventional U-Net network adopts a BN (batch normalization) layer, which is beneficial to maintaining the stable distribution of gradients in the network and accelerating the convergence speed of a model, but also changes the statistical correlation among characteristic information, thereby causing the performance reduction of super-resolution; secondly, the traditional U-Net network adopts deconvolution to perform up-sampling coding, so that the images are uneven and the quality of the images obtained by super-resolution is also influenced; in the embodiment, in order to maintain the statistical correlation between the features and the pixels, the BN layer is deleted; in order to obtain better high-frequency details, deconvolution is replaced by sub-pixel convolution, and the reconstruction quality of the super-resolution reconstruction image is greatly improved.
The residual error dense blocks in the residual error dense network unit are used for extracting local features, the local features extracted by the previous residual error dense block are directly connected with the current residual error dense block, so that the information flow is ensured to be communicated, and the local features extracted by each residual error dense block are fused by a fusion layer and are fully utilized; the spatial attention network unit endows the interpolation image with different weights to the features of different spatial positions, and fuses the feature weight image and the local fusion features, so that the local fusion features are distinguished in importance, and high-frequency information is fully utilized; the global jump connection unit directly performs up-sampling on the interpolation image, avoids loss of partial data, and performs addition operation on the up-sampled image and the partial weight fusion feature at an addition point, so that the reconstruction quality of the initial reconstruction image is more excellent.
The shallow layer features extracted by the residual dense network unit are generally output from the first convolutional layer and are connected to the fused residual feature layer (the summing point) through jumping, but the first convolutional layer can cause loss of partial feature learning, so that before the feature learning is modified to the input of the first convolutional layer, the LR image (interpolation image) is directly up-sampled through the global jumping connection unit and is added to the summing point, and the loss of partial feature learning is reduced;
in the training and testing of the reconstruction system provided by the embodiment, data sets widely used in the super-resolution field are adopted, including DIV2K, Set5, Set14, BSDS100 and Urban100, wherein DIV2K is used for training and verification, and the rest is used for testing. In the super-resolution performance evaluation, PSNR and SSIM are used as evaluation indexes. The two evaluation indexes are widely applied to the evaluation of image reconstruction quality, and the higher the PSNR and SSIM are, the better the reconstructed image quality is;
as shown in the following table, the color image encryption compression and super-resolution reconstruction system provided by the embodiment tests data on the data sets Set5, Set14, BSDS100 and Urban100 under different down-sampling factors s, wherein CR represents the compression ratio, Max, Min and Average represent the maximum value, minimum value and Average value of the compression ratio respectively, and Diff represents the error file E attached theretoimgThe increment of the post-compression ratio, σ, represents the corresponding standard deviation:
the color image encryption compression and super-resolution reconstruction system provided in this embodiment is denoted as rdsn (ors), which is shown in the following table, when the down-sampling factor s is 2, compared with PSNR and SSIM indexes tested by conventional reconstruction methods Bicubic, a +, SRCNN, VDSR, MemNet, ESPCN, SRDenseNet, laprn, EDSR, RDN, cann, CSAR, SAN, and DRLN on data sets Set5, Set14, BSDS100, and Urban 100:
it can be seen that, when the down-sampling factor s is 2, the color image encryption compression and super-resolution reconstruction system provided by this embodiment obtains the highest PSNR and SSIM indexes on the data sets Set5, Set14, BSDS100, and Urban100, and the reconstructed image quality is the best;
as shown in the following table, the comparison between the PSNR and SSIM indexes obtained by testing different data sets with the conventional reconstruction method when the down-sampling factor s is 3 is rdsn (outer):
it can be seen that, when the down-sampling factor s is 3, the color image encryption compression and super-resolution reconstruction system provided by this embodiment obtains the highest PSNR and SSIM indexes on the data sets Set5, Set14 and BSDS100, and the reconstructed image quality is the best;
as shown in the following table, the comparison between the PSNR and SSIM indexes obtained by testing different data sets with the conventional reconstruction method is performed when rdsn (outer) is down-sampled by a factor s ═ 4:
it can be seen that, when the down-sampling factor s is 4, the color image encryption compression and super-resolution reconstruction system provided by this embodiment obtains the highest PSNR and SSIM indexes on the data sets Set5, Set14 and BSDS100, and the reconstructed image quality is the best;
in summary, when the down-sampling factor s is 2-4, the PSNR and SSIM indexes obtained by the color image encryption compression and super-resolution reconstruction system provided by this embodiment through testing on the data sets Set5, Set14, BSDS100, and Urban100 are substantially the highest, and the reconstructed image quality is the best.
ZNNR and DNSR are common unsupervised super-resolution models, and as shown in the following table, when the down-sampling factor s is 2-4, the color image encryption compression and super-resolution reconstruction system provided in this embodiment tests the obtained PSNR and SSIM indexes and the comparative data of the ZNNR and DNSR models on different data sets:
it can be seen that, when the down-sampling factor s is 2-4, the color image encryption compression and super-resolution reconstruction system provided by the embodiment tests the PSNR and SSIM indexes obtained by the data sets Set5, Set14, BSDS100 and Urban100 to be substantially the highest, and the reconstructed image quality is the best.
Example 2
The embodiment provides a color image encryption compression and hyper-division reconstruction method, as shown in fig. 8, the method includes:
s1: preprocessing an original image to obtain a coding sequence;
s2: carrying out encryption and compression operations on the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence;
s3: carrying out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image;
s4: obtaining local features, feature weight graphs and up-sampling images of the interpolation images, fusing to generate an initial reconstruction image, correcting the initial reconstruction image by using the decryption image, and outputting a final reconstruction image.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A color image encryption compression and super-resolution reconstruction system is characterized by comprising a preprocessing module, an encryption compression module, a decryption module and a reconstruction module;
the method comprises the steps that an original image is input into a preprocessing module, the preprocessing module carries out preprocessing operation on the original image to obtain a coding sequence, and then the original image and the coding sequence are input into an encryption compression module; the encryption compression module encrypts and compresses the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence, and sends the encrypted compressed image file and the encrypted coding sequence to the decryption module; the decryption module carries out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image, and sends the interpolation image and the decrypted image to the reconstruction module; the reconstruction module acquires local features, a feature weight graph and an up-sampling image of the interpolation image, fuses to generate an initial reconstruction image, corrects the initial reconstruction image by using the decryption image, and outputs a final reconstruction image.
2. The system for color image encryption compression and super-resolution reconstruction according to claim 1, wherein the preprocessing module performs preprocessing operations on the original image specifically as follows:
the preprocessing module is used for input original image IorPerforming interpolation operation and downsampling operation to obtain interpolated image IbicAnd downsampling of subgraph Iordwn；
Pixel-by-pixel computed interpolation image IbicWith down-sampled subgraph IordwnError file E ofimgFor error file EimgCoding to obtain coding sequence Eac；
Coding sequence EacAttached to the original image IorAnd after the end, sending the data to an encryption compression module.
3. The color image encryption compression and super-resolution reconstruction system according to claim 2, wherein the encryption compression module comprises an encryption unit and a compression unit;
the encryption unit uses the encryption key to encrypt the received original image IorAnd coding sequence EacCarrying out encryption operation to obtain an encrypted image file IencAnd an encrypted coding sequence EencTo encrypt an image file IencAnd an encrypted coding sequence EencSending the key to a compression unit, and sending an encryption key used for encryption to a decryption module through a secure channel;
the compression unit encrypts the image file IencCompressing to obtain a compressed image file Idwn(ii) a To compress an image file IdwnAnd an encrypted coding sequence EencAnd sending the data to a decryption module through a common channel.
4. The color image encryption compression and super-resolution reconstruction system according to claim 3, wherein the decryption module performs decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain the interpolated image and the decrypted image by the specific method comprising:
the decryption module is used for decrypting the compressed image file I according to the received encryption keydwnAnd an encrypted coding sequence EencPerforming decryption operation to obtain decrypted image IdecAnd coding sequence Eac；
For coding sequence EacDecoding to obtain an error file Eimg；
Will decrypt the image IdecAnd an error file EimgAdding pixel by pixel to obtain an interpolation image Ibic。
5. The color image encryption compression and super-resolution reconstruction system according to claim 4, wherein the reconstruction module comprises a residual error intensive network unit, a global jump connection unit, a spatial attention network unit and a modification construction unit;
the space attention network unit endows different weights to the features of the interpolation image at different space positions to obtain a feature weight graph, and the feature weight graph is output to the residual error dense network unit;
the global jump connection unit is used for up-sampling the interpolation image to obtain an up-sampled image, and jump-connecting the up-sampled image to the residual error dense network unit;
the residual dense network unit is used for extracting local features of the interpolation image and fusing the local features and the feature weight graph into local weight fusion features; adding the local weight fusion features and the up-sampled image to generate an initial reconstruction image, and outputting the initial reconstruction image to a correction construction unit;
the correction construction unit corrects the initial reconstructed image by using the decrypted image to obtain a final reconstructed image.
6. The color image encryption compression and super-resolution reconstruction system according to claim 5, wherein the spatial attention network unit is modified based on a conventional U-Net network, comprising: the BN layer in the conventional U-Net network is removed and the deconvolution replaced with a sub-pixel convolution.
7. The color image encryption compression and super-resolution reconstruction system according to claim 6, wherein the residual dense network unit comprises a first convolutional layer, a second convolutional layer, D residual dense blocks, a fusion layer, a 1 x 1 convolutional layer, a third convolutional layer, a fusion point, an upsampling layer, a summing point and a fourth convolutional layer which are connected in sequence;
extracting shallow layer characteristics from the input interpolation image through a first convolution layer and a second convolution layer; d residual error dense blocks respectively extract the features of each layer, and each residual error dense block sends one extracted local feature to the fusion layer for local feature fusion; after the number of channels is reduced by 1 multiplied by 1 convolution and local fusion characteristics are further extracted by a third convolution layer, fusing the fusion points with the characteristic weight graph extracted by the space attention network unit to obtain local weight fusion characteristics; after the up-sampling layer is amplified, the up-sampling image obtained by the global jump connection unit is added at an addition point, and an initial reconstruction image is output through a fourth convolution layer.
8. The color image encryption compression and super-resolution reconstruction system according to claim 2 or 4, wherein the error file E is encoded using arithmetic codingimgCoding to obtain coding sequence Eac(ii) a Coding sequence E using arithmetic codingacDecoding to obtain an error file Eimg。
9. The color image encryption compression and super-resolution reconstruction system according to claim 3, wherein the encryption unit uses a stream cipher-based encryption method for the received original image IorAnd coding sequence EacCarrying out encryption operation; the compression unit encrypts the image file I using uniform downsamplingencCompression is performed.
10. A color image encryption compression and super-division reconstruction method, the method comprising:
s1: preprocessing an original image to obtain a coding sequence;
s2: carrying out encryption and compression operations on the original image and the coding sequence to obtain an encrypted compressed image file and an encrypted coding sequence;
s3: carrying out decryption operation on the encrypted compressed image file and the encrypted coding sequence to obtain an interpolation image and a decrypted image;
s4: obtaining local features, feature weight graphs and up-sampling images of the interpolation images, fusing to generate an initial reconstruction image, correcting the initial reconstruction image by using the decryption image, and outputting a final reconstruction image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110781008.6A CN113660386B (en) | 2021-07-09 | 2021-07-09 | Color image encryption compression and super-resolution reconstruction system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110781008.6A CN113660386B (en) | 2021-07-09 | 2021-07-09 | Color image encryption compression and super-resolution reconstruction system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113660386A true CN113660386A (en) | 2021-11-16 |
| CN113660386B CN113660386B (en) | 2022-04-26 |
Family
ID=78477226
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110781008.6A Active CN113660386B (en) | 2021-07-09 | 2021-07-09 | Color image encryption compression and super-resolution reconstruction system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113660386B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114222123A (en) * | 2021-12-15 | 2022-03-22 | 华南农业大学 | System and method for lossy compression and reconstruction of encrypted image with any compression rate |
| CN115471206A (en) * | 2022-09-29 | 2022-12-13 | 深圳标普云科技有限公司 | Contract management and control method and contract management system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111464718A (en) * | 2020-03-31 | 2020-07-28 | 孙甲飞 | Encryption image reversible information hiding method based on interpolation technology |
| CN111681166A (en) * | 2020-06-02 | 2020-09-18 | 重庆理工大学 | Image super-resolution reconstruction method of stacked attention mechanism coding and decoding unit |
| US10909728B1 (en) * | 2019-05-01 | 2021-02-02 | Amazon Technologies, Inc. | Learned lossy image compression codec |
-
2021
- 2021-07-09 CN CN202110781008.6A patent/CN113660386B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10909728B1 (en) * | 2019-05-01 | 2021-02-02 | Amazon Technologies, Inc. | Learned lossy image compression codec |
| CN111464718A (en) * | 2020-03-31 | 2020-07-28 | 孙甲飞 | Encryption image reversible information hiding method based on interpolation technology |
| CN111681166A (en) * | 2020-06-02 | 2020-09-18 | 重庆理工大学 | Image super-resolution reconstruction method of stacked attention mechanism coding and decoding unit |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114222123A (en) * | 2021-12-15 | 2022-03-22 | 华南农业大学 | System and method for lossy compression and reconstruction of encrypted image with any compression rate |
| CN115471206A (en) * | 2022-09-29 | 2022-12-13 | 深圳标普云科技有限公司 | Contract management and control method and contract management system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113660386B (en) | 2022-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113660386B (en) | Color image encryption compression and super-resolution reconstruction system and method | |
| US7400764B2 (en) | Compression and decompression of media data | |
| CN104284190B (en) | Compressed image steganography encoding method based on AMBTC high and low average optimization | |
| CN103607591A (en) | Image compression method combining super-resolution reconstruction | |
| WO2021047471A1 (en) | Image steganography method and apparatus, and image extraction method and apparatus, and electronic device | |
| Yao et al. | Dual-JPEG-image reversible data hiding | |
| CN105791854A (en) | Singular value modification video steganographic algorithm based on combination with improved matrix coding | |
| CN110445949A (en) | The domain AMBTC reversible information hidden method based on histogram displacement | |
| CN112911290A (en) | Ciphertext image reversible data hiding method based on predictive difference compression coding | |
| Rajalakshmi et al. | ZLBM: zero level binary mapping technique for video security | |
| CN112132738A (en) | Image robust steganography method with reference generation | |
| CN108521535B (en) | A kind of Information hiding transmission method based on image blend processing | |
| Yin et al. | Reversible data hiding in binary images by flipping pattern pair with opposite center pixel | |
| CN103581677B (en) | Image encrypted compression and decompression decryption method and device | |
| Ren et al. | Reversible data hiding scheme in encrypted images based on homomorphic encryption and pixel value ordering | |
| CN114782563B (en) | Secret image sharing method and system for JPEG image | |
| CN115131253B (en) | Secret image sharing method and system for resisting JPEG recompression | |
| Wang et al. | A customized deep network based encryption-then-lossy-compression scheme of color images achieving arbitrary compression ratios | |
| CN114222123B (en) | System and method for lossy compression and reconstruction of encrypted image with any compression rate | |
| Yang et al. | Graph-convolution network for image compression | |
| CN105046633B (en) | A kind of image lossless transform method | |
| CN107135391A (en) | A kind of video-aware hash algorithm for H.264 compression domain | |
| Chen et al. | Classification-based spatial error concealment for visual communications | |
| Peng et al. | An optimized algorithm based on generalized difference expansion method used for HEVC reversible video information hiding | |
| Zhang et al. | Dual-layer image compression via adaptive downsampling and spatially varying upconversion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |