WO2018120724A1 - 图像篡改检测方法、系统、电子装置及存储介质 - Google Patents

图像篡改检测方法、系统、电子装置及存储介质 Download PDF

Info

Publication number
WO2018120724A1
WO2018120724A1 PCT/CN2017/091312 CN2017091312W WO2018120724A1 WO 2018120724 A1 WO2018120724 A1 WO 2018120724A1 CN 2017091312 W CN2017091312 W CN 2017091312W WO 2018120724 A1 WO2018120724 A1 WO 2018120724A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
tampering
detection result
detected
feature
Prior art date
Application number
PCT/CN2017/091312
Other languages
English (en)
French (fr)
Inventor
王健宗
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Priority to AU2017389535A priority Critical patent/AU2017389535B2/en
Priority to SG11201808824YA priority patent/SG11201808824YA/en
Priority to EP17882272.2A priority patent/EP3396625A4/en
Priority to KR1020187017248A priority patent/KR102168397B1/ko
Priority to JP2018528057A priority patent/JP2019503530A/ja
Priority to US16/084,227 priority patent/US10692218B2/en
Publication of WO2018120724A1 publication Critical patent/WO2018120724A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/0021—Image watermarking
    • G06T1/0028—Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2201/00—General purpose image data processing
    • G06T2201/005—Image watermarking
    • G06T2201/0201—Image watermarking whereby only tamper or origin are detected and no embedding takes place
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • the present invention relates to the field of computer technologies, and in particular, to an image tamper detecting method, system, electronic device, and storage medium.
  • the existing image tamper detection technology mostly detects a single type of tampering, for example, can only detect a single type of tampering such as copying, moving or splicing; and, because most existing tamper detection algorithms work based on JPEG compression traces, so only JPEG type image tamper detection can be processed, so that the image format capable of image tamper detection is relatively simple.
  • the existing image tampering detection algorithm needs to rely on complex feature extraction, and the extracted features are not robust to the tamper type changes, which leads to the inaccuracy of image tamper detection.
  • the main object of the present invention is to provide an image tamper detecting method, system, electronic device and storage medium, which are intended to accurately detect image tampering of different types and formats.
  • a first aspect of the present application provides an image tampering detection method, the method comprising the following steps:
  • the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature, and the tamper detection result corresponding to the to-be-detected image is determined according to the generated complex tampering feature, and the tampering detection result includes tampering And not tampering.
  • the second aspect of the present application provides an image tampering detection system, where the image tampering detection system includes:
  • An extraction module configured to perform block segmentation on the image to be detected, divide the image to be detected into a plurality of image small fragments, and extract an initial tamper detection feature from each image small fragment;
  • a detecting module configured to encode the extracted initial tamper detecting feature by using a predetermined encoder to generate a complex tampering feature, and determine a tampering detection result corresponding to the to-be-detected image according to the generated complex tampering feature, the tampering detection result Including tampering and Not tampering.
  • a third aspect of the present application provides an electronic device, including a processing device and a storage device; the storage device stores an image tampering detection program including at least one computer readable instruction, the at least one computer readable instruction being The device executes to do the following:
  • the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature, and the tamper detection result corresponding to the to-be-detected image is determined according to the generated complex tampering feature, and the tampering detection result includes tampering And not tampering.
  • a fourth aspect of the present application provides a computer readable storage medium having stored thereon at least one computer readable instruction executable by a processing device to:
  • the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature, and the tamper detection result corresponding to the to-be-detected image is determined according to the generated complex tampering feature, and the tampering detection result includes tampering And not tampering.
  • the image tampering detection method, system, electronic device and storage medium divides the image to be detected into several small image fragments, extracts initial tamper detection features from each image small fragment, and uses a predetermined encoder pair
  • the extracted initial tamper detecting feature is encoded to generate a complex tampering feature, and the tampering detection result of the image to be detected is determined according to the complex tampering feature. Since the image to be detected is segmented, the initial tamper detection feature based on each image small fragment and the tampering of the image to be detected by the encoder using the complex tampering feature generated by the initial tamper detection feature encoding are not dependent on JPEG compression.
  • the type of image that can be tamper-detected is not limited to JPEG, and detection by means of segmentation and encoding can accurately detect various types of image tampering modes, thereby realizing different types and formats. Image tampering for accurate detection.
  • FIG. 1 is a schematic diagram of an application environment of a preferred embodiment of an image tamper detecting method according to the present invention
  • FIG. 2 is a schematic flow chart of an embodiment of an image tampering detecting method according to the present invention
  • step S10 in FIG. 2 is a schematic diagram of a refinement process of step S10 in FIG. 2;
  • FIG. 4 is a schematic structural diagram of a stacking automatic encoder in an embodiment of an image tamper detecting method according to the present invention
  • FIG. 5 is a schematic diagram of functional modules of an embodiment of an image tamper detecting system according to the present invention.
  • FIG. 6 is a schematic diagram of a refinement function module of the extraction module 01 of FIG. 5.
  • the electronic device 1 is an apparatus capable of automatically performing numerical calculation and/or information processing in accordance with an instruction set or stored in advance.
  • the electronic device 1 may be a computer, a single network server, a server group composed of multiple network servers, or a cloud-based cloud composed of a large number of hosts or network servers, where cloud computing is a type of distributed computing.
  • a super virtual computer consisting of a group of loosely coupled computers.
  • the electronic device 1 includes a storage device 11, a processing device 12, and the like.
  • the processing device 12 is for supporting the operation of the electronic device 1, which may include one or more microprocessors, digital processors, and the like.
  • the storage device 11 is for storing various data and computer readable instructions, which may include one or more non-volatile memories such as a ROM, an EPROM or a Flash Memory.
  • the storage device 11 stores an image tampering detection program including at least one computer readable instruction stored in the storage device 11, the at least one computer readable instruction being executable by the processor device 12 to implement Image tampering detection method of each embodiment of the present application.
  • the invention provides an image tampering detection method.
  • FIG. 2 is a schematic flowchart diagram of an embodiment of an image tampering detecting method according to the present invention.
  • the image tampering detection method comprises:
  • Step S10 Perform block segmentation on the image to be detected, divide the image to be detected into several small image fragments, and extract initial tamper detection features from each image small fragment.
  • block-dividing is performed on various types of to-be-detected images, such as JPEG, PNG, and GIF, and the image to be detected is divided into small pieces of images of N*M (for example, 32*32), and N and M are A positive integer, and an initial tamper detection feature is extracted from each small image fragment.
  • the initial tamper detection feature may include a color feature, a pixel feature, and the like, which are not limited herein.
  • the image to be detected when the image to be detected is subjected to block segmentation, the image to be detected may be equally divided into small pieces of images of the same size, or the image to be detected may be determined according to a certain Proportional or random is divided into small fragments of different sizes, which are not limited here.
  • the small fragments of the image may be squares, rectangles, and the like having a regular shape, or may be irregularly shaped fragments, which are not limited herein.
  • step S20 the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature, and the tampering detection result corresponding to the image to be detected is determined according to the generated complex tampering feature, and the tampering detection result includes Tampering and not tampering.
  • the extracted initial tamper detecting feature is encoded by using a predetermined encoder to generate a complex tampering feature, and finally detecting according to the generated complex tampering feature to determine the to-be-detected
  • the image is a state that has been tampered with or not tampered with.
  • the predetermined encoder may be a relatively fixed encoder, or may be an encoder capable of deep learning such as learning, training, etc., for example, may be a single automatic encoder, a stack automatic encoder, etc., do not do here. limited.
  • the predetermined encoder is an encoder capable of deep learning such as learning, training, and the like, such as a stack automatic encoder.
  • the training process of the predetermined encoder is as follows:
  • D. Perform block segmentation on each image sample, and divide each image sample into N*M (for example, 32*32) image small fragments, N and M are positive integers, and extract from each image small fragment of each image sample.
  • N*M for example, 32*32
  • M are positive integers
  • All image samples are divided into a first data set and a second data set according to a ratio of X:Y (for example, 8:2), and the number of image samples in the first data set is greater than the number of image samples in the second data set, first The data set is used as a training set, and the second data set is used as a test set, wherein X is greater than 0 and Y is greater than 0;
  • the training is ended, or if the accuracy of the trained encoder is less than the preset accuracy, the number of image samples is increased. And repeat the above steps D, E and F until the accuracy of the trained encoder is greater than or equal to the preset accuracy.
  • the image to be detected is divided into a plurality of small fragments of the image, the initial tamper detection feature is extracted from each small image fragment, and the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature.
  • the tampering feature determines a tampering detection result of the image to be detected. Since the image to be detected is segmented, the initial tamper detection feature based on each image small fragment and the tampering of the image to be detected by the encoder using the complex tampering feature generated by the initial tamper detection feature encoding are not dependent on JPEG compression.
  • the type of image that can be tamper-detected is not limited to JPEG, and detection by means of segmentation and encoding can accurately detect various types of image tampering modes, thereby realizing different types and formats. Image tampering for accurate detection.
  • step S10 may include:
  • Step S101 converting the image to be detected from a first color space (for example, an RGB color space) to a second color space (for example, a YCrCb color space);
  • a first color space for example, an RGB color space
  • a second color space for example, a YCrCb color space
  • Step S102 performing block segmentation on the image to be detected, and dividing the image to be detected into small image fragments of N*M (for example, 32*32), where N and M are positive integers;
  • Step S103 applying a preset wavelet function, such as a plurality of levels (for example, three levels) of two-dimensional Doubeches Wavelet decomposition, on each second color space component on each image small fragment, to Obtaining a plurality of (eg, 30) corresponding pixel coefficient maps (Coefficient Maps), and calculating summary statistical coefficients for each of the pixel coefficient maps (eg, pixel standard deviation, pixel mean, and/or pixel corresponding to each pixel coefficient map) Sum), and apply wavelet functions such as Daubechies Orthogonal wavelets D2-D5 to obtain multiple summary statistical coefficients on each image small fragment, each of which is the initial of the corresponding image small fragment Tamper detection features.
  • a preset wavelet function such as a plurality of levels (for example, three levels) of two-dimensional Doubeches Wavelet decomposition
  • the image to be detected is divided into a plurality of small pieces of the image for feature extraction, and each detail feature of the image to be detected can be extracted in a more detailed manner, and the Dobecy wavelet and Dobecy are utilized.
  • the orthogonal statistical method is used to obtain the summary statistical coefficient of each image small fragment as the initial tamper detection feature, which can make the extracted initial tamper detection feature more accurate.
  • the predetermined encoder is a Stacked Auto-Encoder (SAE).
  • SAE Stacked Auto-Encoder
  • FIG. 4 is a stacking method of an image tamper detecting method according to an embodiment of the present invention. Schematic diagram of the structure of the automatic encoder.
  • the stack autoencoder is a stack of multiple basic autoencoders, and the output of each layer of autoencoder is the input of the next layer of autoencoder.
  • the stack auto-encoder also includes a Multi-Layer Perceptron (MLP), which interfacing with the last layer of the auto-encoder to determine the complex tampering features generated The tampering detection result corresponding to the image to be detected is performed.
  • MLP Multi-Layer Perceptron
  • W (l) and b (l) represent the weight matrix and offset vector of the l-th layer autoencoder neuron.
  • the input of the l-th layer autoencoder is Z (l) and the output is ae. (l) , then:
  • f(.) is the activation function of the neuron.
  • sigmoid function is used, and its expression is as follows:
  • the decoding process of SAE is the inverse of the encoding process.
  • the relevant input and output expressions are as follows:
  • n the total number of layers in the coding layer. It can be seen from the above formula that the number of layers in the coding layer in the SAE is the same as the number of layers in the decoding layer, but only the coding layer is used in the test. Therefore, when using SAE to extract features from the input, the final feature representation is the activation value vector Y of the last layer, with the following expression:
  • the input layer of the stacking autoencoder receives the initial tamper detection feature, and then passes the received initial tamper detection feature to the intermediate implicit coding layer for feature dimensionality reduction filtering, and then reconstructs the decoded layer to reconstruct the original
  • the input layer by calculating the difference between the reconstructed input and the original input, guides the training of the coding and decoding layers of the SAE, and finally ensures that the final output of the SAE coding layer is a high-level complex tampering feature of the original input features, using the advanced Complex tampering features are beneficial to improve the classification and classification ability of SAE extracted features, and thus improve the accuracy of image tampering detection.
  • step S20 may include:
  • the initial tamper detecting feature of each image small fragment is encoded by a predetermined encoder to generate a complex tampering feature of each image small fragment;
  • the tampering detection result corresponding to the small image fragment is falsified, it is determined that the tampering detection result corresponding to the image to be detected is falsified.
  • the complex tampering feature of each image small fragment may be used. Detect whether the small fragments of each image are falsified. If the first tampering detection result corresponding to one or more image small fragments is falsified, indicating that the image to be detected has a falsified portion, it is determined that the tampering detection result corresponding to the image to be detected is falsified.
  • the tampering detection result corresponding to all the small fragments of the image is not falsified, the part of the image to be detected that has not been tampered with, Then, it is determined that the tampering detection result corresponding to the image to be detected is not tampering.
  • the stacking autoencoder further includes an adjacent area sensing layer for determining a tampering detection result corresponding to the image to be detected according to the scene information of the adjacent area, and the step S20 is further performed.
  • the second tampering detection result corresponding to the small image fragments is not falsified, and the first tampering detection result corresponding to the small image fragment is falsified, it is determined that the tampering detection result corresponding to the image to be detected is not tampering.
  • the neighboring area sensing layer is also used to calculate the adjacent scene information of each image small fragment, and is determined according to the adjacent scene information of each image small fragment.
  • the second tampering detection result corresponding to each image small fragment.
  • the tampering detection of the image to be detected is performed on the basis of comprehensively considering the first tampering detection result and the second tampering detection result.
  • the second tampering detection result corresponding to all the small pieces of the image is not falsified, and the first tampering detection result corresponding to the small image fragment is falsified, determining that the tampering detection result corresponding to the image to be detected is not tampering, preventing The tampering detection error occurs in the image to be detected due to the misjudgment of the first tampering detection result, etc., which further improves the accuracy and stability of the image tampering detection.
  • the preset algorithm includes:
  • N(p) For an image small fragment P, assuming its neighboring scene information is N(p), the expression of N(p) is as follows:
  • the MLP classification result indicating the i-th adjacent image small fragment of the small fragment P of the image, K can be set to 8 (ie, 8 possible directions of each image small fragment are used as the final tampering determination classification) If the average value of the classification result corresponding to the small fragment P of the image is greater than or equal to a preset threshold (for example, 0.5), it is determined that the second tampering detection result corresponding to the small fragment P of the image is a falsified result, or if the image is small If the average value of the classification result corresponding to the fragment P is less than the preset threshold, it is determined that the second tampering detection result corresponding to the current image small fragment P is an untampered result.
  • a preset threshold for example, 0.5
  • the invention further provides an image tamper detecting system.
  • FIG. 5 is a schematic diagram of functional modules of an embodiment of an image tamper detecting system according to the present invention.
  • the image tampering detection system comprises:
  • the extraction module 01 is configured to perform block segmentation on the image to be detected, divide the image to be detected into several small image fragments, and extract an initial tamper detection feature from each image small fragment.
  • block-dividing is performed on various types of to-be-detected images, such as JPEG, PNG, and GIF, and the image to be detected is divided into small pieces of images of N*M (for example, 32*32), and N and M are A positive integer, and an initial tamper detection feature is extracted from each small image fragment.
  • the initial tamper detection feature may include a color feature, a pixel feature, and the like, which are not limited herein.
  • the image to be detected may be equally divided into small pieces of image of the same size, or the image to be detected may be divided into small pieces of images of different sizes according to a certain ratio or random. Not limited.
  • the small fragments of the image may be squares, rectangles, and the like having a regular shape, or may be irregularly shaped fragments, which are not limited herein.
  • the detecting module 02 is configured to encode the extracted initial tamper detecting feature by using a predetermined encoder to generate a complex tampering feature, and determine a tampering detection result corresponding to the to-be-detected image according to the generated complex tampering feature, the tamper detecting
  • the results include falsification and tampering.
  • the extracted initial tamper detecting feature is encoded by using a predetermined encoder to generate a complex tampering feature, and finally detecting according to the generated complex tampering feature to determine the to-be-detected
  • the image is a state that has been tampered with or not tampered with.
  • the predetermined encoder may be a relatively fixed encoder, or may be an encoder capable of deep learning such as learning, training, etc., for example, may be a single automatic encoder, a stack automatic encoder, etc., do not do here. limited.
  • the predetermined encoder is an encoder capable of deep learning such as learning, training, and the like, such as a stack automatic encoder.
  • the training process of the predetermined encoder is as follows:
  • All image samples are divided into a first data set and a second data set according to a ratio of X:Y (for example, 8:2), and the number of image samples in the first data set is greater than the number of image samples in the second data set, first The data set is used as a training set, and the second data set is used as a test set, wherein X is greater than 0 and Y is greater than 0;
  • the training is ended, or if the accuracy of the trained encoder is less than the preset accuracy, the number of image samples is increased. And repeat the above steps D, E and F until the accuracy of the trained encoder is greater than or equal to the preset accuracy.
  • the image to be detected is divided into a plurality of small fragments of the image, the initial tamper detection feature is extracted from each small image fragment, and the extracted initial tamper detection feature is encoded by using a predetermined encoder to generate a complex tampering feature. Determining the tampering detection result of the image to be detected according to the complex tampering feature. Since the image to be detected is segmented, the initial tamper detection feature based on each image small fragment and the tampering of the image to be detected by the encoder using the complex tampering feature generated by the initial tamper detection feature encoding are not dependent on JPEG compression.
  • the type of image that can be tamper-detected is not limited to JPEG, and detection by means of segmentation and encoding can accurately detect various types of image tampering modes, thereby realizing different types and formats. Image tampering for accurate detection.
  • the foregoing extraction module 01 may include:
  • a converting unit 011 configured to convert the image to be detected from a first color space (for example, an RGB color space) to a second color space (for example, a YCrCb color space);
  • a dividing unit 012 configured to perform block segmentation on the image to be detected, and divide the image to be detected into small image fragments of N*M (for example, 32*32), where N and M are positive integers;
  • the obtaining unit 013 is configured to apply a preset wavelet function such as a plurality of levels (for example, three levels) of two-dimensional Doubeches Wavelet on each second color space component on each image small fragment.
  • a preset wavelet function such as a plurality of levels (for example, three levels) of two-dimensional Doubeches Wavelet on each second color space component on each image small fragment.
  • a preset wavelet function such as a plurality of levels (for example, three levels) of two-dimensional Doube
  • the image to be detected is divided into a plurality of small pieces of the image for feature extraction, and each detail feature of the image to be detected can be extracted in a more detailed manner, and the Dobecy wavelet and Dobecy are utilized.
  • the orthogonal statistical method is used to obtain the summary statistical coefficient of each image small fragment as the initial tamper detection feature, which can make the extracted initial tamper detection feature more accurate.
  • the predetermined encoder is a Stacked Auto-Encoder (SAE).
  • SAE Stacked Auto-Encoder
  • the stack auto-encoder is layered by a plurality of basic auto-encoders. The output of each layer of the automatic encoder is the input of the next layer of automatic encoder.
  • the stack auto-encoder also includes a Multi-Layer Perceptron (MLP), which interfacing with the last layer of the auto-encoder to determine the complex tampering features generated The tampering detection result corresponding to the image to be detected is performed. For example, suppose W (l) and b (l) represent the weight matrix and offset vector of the l-th layer autoencoder neuron.
  • the input of the l-th layer autoencoder is Z (l) and the output is ae. (l) , then:
  • f(.) is the activation function of the neuron.
  • sigmoid function is used, and its expression is as follows:
  • the decoding process of SAE is the inverse of the encoding process.
  • the relevant input and output expressions are as follows:
  • n the total number of layers in the coding layer. It can be seen from the above formula that the number of layers in the coding layer in the SAE is the same as the number of layers in the decoding layer, but only the coding layer is used in the test. Therefore, when using SAE to extract features from the input, the final feature representation is the activation value vector Y of the last layer, with the following expression:
  • the input layer of the stacking autoencoder receives the initial tamper detection feature, and then passes the received initial tamper detection feature to the intermediate implicit coding layer for feature dimensionality reduction filtering, and then reconstructs the decoded layer to reconstruct the original
  • the input layer by calculating the difference between the reconstructed input and the original input, guides the training of the coding and decoding layers of the SAE, and finally ensures that the final output of the SAE coding layer is a high-level complex tampering feature of the original input features, using the advanced Complex tampering features are beneficial to improve the classification and classification of SAE extracted features Force, thereby improving the accuracy of image tamper detection.
  • the foregoing detecting module 02 can also be used to:
  • the initial tamper detecting feature of each image small fragment is encoded by a predetermined encoder to generate a complex tampering feature of each image small fragment;
  • the tampering detection result corresponding to the small image fragment is falsified, it is determined that the tampering detection result corresponding to the image to be detected is falsified.
  • the complex tampering feature of each image small fragment may be used. Detect whether the small fragments of each image are falsified. If the first tampering detection result corresponding to one or more image small fragments is falsified, indicating that the image to be detected has a falsified portion, it is determined that the tampering detection result corresponding to the image to be detected is falsified. If the first tamper detection result corresponding to all the small image fragments is not falsified, the tamper detection result corresponding to the to-be-detected image is determined to be untampered.
  • a predetermined encoder such as a stack auto-encoder to generate a corresponding complex tampering feature
  • the stacking autoencoder further includes an adjacent area sensing layer for determining a tampering detection result corresponding to the image to be detected according to the scene information of the adjacent area, the detecting module 02 Can also be used
  • the second tampering detection result corresponding to the small image fragments is not falsified, and the first tampering detection result corresponding to the small image fragment is falsified, it is determined that the tampering detection result corresponding to the image to be detected is not tampering.
  • the neighboring area sensing layer is also used to calculate the adjacent scene information of each image small fragment, and is determined according to the adjacent scene information of each image small fragment.
  • the second tampering detection result corresponding to each image small fragment.
  • the tampering detection of the image to be detected is performed on the basis of comprehensively considering the first tampering detection result and the second tampering detection result.
  • the second tampering detection result corresponding to all the small pieces of the image is not falsified, and the first tampering detection result corresponding to the small image fragment is falsified, determining that the tampering detection result corresponding to the image to be detected is not tampering, preventing The image to be detected is caused by a misjudgment of the first tampering detection result or the like The occurrence of tampering detection errors occurs, which further improves the accuracy and stability of image tampering detection.
  • the preset algorithm includes:
  • N(p) For an image small fragment P, assuming its neighboring scene information is N(p), the expression of N(p) is as follows:
  • the MLP classification result indicating the i-th adjacent image small fragment of the small fragment P of the image, K can be set to 8 (ie, 8 possible directions of each image small fragment are used as the final tampering determination classification) If the average value of the classification result corresponding to the small fragment P of the image is greater than or equal to a preset threshold (for example, 0.5), it is determined that the second tampering detection result corresponding to the small fragment P of the image is a falsified result, or if the image is small If the average value of the classification result corresponding to the fragment P is less than the preset threshold, it is determined that the second tampering detection result corresponding to the current image small fragment P is an untampered result.
  • a preset threshold for example, 0.5
  • the foregoing extraction module 01, the detection module 02, and the like may be embedded in or independent of the electronic device in hardware, or may be stored in a storage device of the electronic device in software, so that the processing device invokes to execute each of the above.
  • the processing device can be a central processing unit (CPU), a microprocessor, a single chip microcomputer, or the like.
  • the present invention also provides a computer readable storage medium storing an image tamper detecting system, the image tamper detecting system being executable by at least one processor to cause the at least one processor.
  • the specific implementation process of the image tampering detection method such as the steps S10, S20, and S30, is as described above, and is not described here.
  • the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and can also be implemented by hardware, but in many cases, the former is A better implementation.
  • the technical solution of the present invention which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk,
  • the optical disc includes a number of instructions for causing a terminal device (which may be a cell phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the methods described in various embodiments of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Storage Device Security (AREA)

Abstract

本发明公开了一种图像篡改检测方法、系统、电子装置及存储介质，该方法包括：A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。本发明实现对不同类型、格式的图像篡改进行准确的检测。

Description

图像篡改检测方法、系统、电子装置及存储介质
优先权申明
本申请基于巴黎公约申明享有2016年12月30日递交的申请号为CN201611265608.2、名称为“图像篡改检测方法及装置”中国专利申请的优先权，该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本发明涉及计算机技术领域，尤其涉及一种图像篡改检测方法、系统、电子装置及存储介质。
背景技术
目前，现有的图像篡改检测技术大都是针对单一类型的篡改进行检测，例如，只能检测复制、移动或者拼接等单一类型的篡改；而且，由于现有大多数篡改检测算法的工作原理是基于JPEG压缩痕迹，所以只能处理JPEG类型的图像篡改检测，使得能够进行图像篡改检测的图像格式比较单一。同时，现有的图像篡改检测算法需要依赖复杂的特征提取，并且提取的特征对篡改类型的变化不具鲁棒性，导致图像篡改检测的准确性得不到保障。
发明内容
本发明的主要目的在于提供一种图像篡改检测方法、系统、电子装置及存储介质，旨在对不同类型、格式的图像篡改进行准确的检测。
本申请第一方面提供一种图像篡改检测方法，所述方法包括以下步骤：
A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
本申请第二方面提供一种图像篡改检测系统，所述图像篡改检测系统包括：
提取模块，用于对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
检测模块，用于利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和 未篡改。
本申请第三方面提供一种电子装置，包括处理设备及存储设备；该存储设备中存储有图像篡改检测程序，其包括至少一个计算机可读指令，该至少一个计算机可读指令可被所述处理设备执行，以实现以下操作：
A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
本申请第四方面提供一种计算机可读存储介质，其上存储有至少一个可被处理设备执行以实现以下操作的计算机可读指令：
A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
本发明提出的图像篡改检测方法、系统、电子装置及存储介质，通过对待检测图像进行块分割成若干图像小碎片，从各个图像小碎片中提取初始篡改检测特征，并利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，根据该复杂篡改特征确定所述待检测图像的篡改检测结果。由于是将待检测图像进行分割后，基于各个图像小碎片的初始篡改检测特征以及利用编码器对该初始篡改检测特征编码生成的复杂篡改特征来对待检测图像的篡改进行检测，并不依赖JPEG压缩痕迹原理，因此，能进行篡改检测的图像类型并不仅限于JPEG，而且，通过分割及编码的方式进行检测，能准确地对多种类型的图像篡改方式进行检测，从而实现对不同类型、格式的图像篡改进行准确的检测。
附图说明
图1为本发明实现图像篡改检测方法的较佳实施例的应用环境示意图；
图2为本发明图像篡改检测方法一实施例的流程示意图；
图3为图2中步骤S10的细化流程示意图；
图4为本发明图像篡改检测方法一实施例中栈化自动编码器的结构示意图；
图5为本发明图像篡改检测系统一实施例的功能模块示意图；
图6为图5中提取模块01的细化功能模块示意图。
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
为了使本发明所要解决的技术问题、技术方案及有益效果更加清楚、明白，以下结合附图和实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
参阅图1所示，是本发明实现图像篡改检测方法的较佳实施例的应用环境示意图。所述电子装置1是一种能够按照事先设定或者存储的指令，自动进行数值计算和/或信息处理的设备。所述电子装置1可以是计算机、也可以是单个网络服务器、多个网络服务器组成的服务器组或者基于云计算的由大量主机或者网络服务器构成的云，其中云计算是分布式计算的一种，由一群松散耦合的计算机集组成的一个超级虚拟计算机。
在本实施例中，电子装置1包括存储设备11及处理设备12等。处理设备12用于支撑电子装置1的运行，其可以包括一个或者多个微处理器、数字处理器等。存储设备11用于存储各种数据及计算机可读指令，其可以包括一个或者多个非易失性存储器，如ROM、EPROM或Flash Memory(快闪存储器)等。在一实施例中，存储设备11中存储有图像篡改检测程序，其包括至少一个存储在存储设备11中的计算机可读指令，该至少一个计算机可读指令可被处理器设备12执行，以实现本申请各实施例的图像篡改检测方法。
本发明提供一种图像篡改检测方法。
参照图2，图2为本发明图像篡改检测方法一实施例的流程示意图。
在一实施例中，该图像篡改检测方法包括：
步骤S10，对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征。
本实施例中，对如JPEG、PNG、GIF等各种类型的待检测图像进行块分割，将所述待检测图像分成N*M(例如，32*32)的图像小碎片，N和M为正整数，并从各个图像小碎片中提取出初始篡改检测特征，该初始篡改检测特征可包括颜色特征、像素特征等，在此不做限定。其中，对待检测图像进行块分割时，可以将所述待检测图像等分成各个相同大小的图像小碎片，也可以将所述待检测图像按一定 比例或随机分成不同大小的图像小碎片，在此不做限定。图像小碎片既可以是形状规则的正方形、长方形等，也可以是形状不规则的碎片，在此不做限定。
步骤S20，利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
提取出各个图像小碎片中的初始篡改检测特征之后，利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征最终进行检测确定所述待检测图像是已篡改或未篡改的状态。其中，该预先确定的编码器可以是相对固定的编码器，也可以是能进行学习、训练等深度学习的编码器，例如可以是单自动编码器、栈式自动编码器等，在此不做限定。
在一种实施方式中，该预先确定的编码器是能进行学习、训练等深度学习的编码器如栈式自动编码器等，该预先确定的编码器的训练过程如下：
C、获取预设数量(例如，10万个)的图像样本，并对各个图像样本的篡改结果进行标识，所述篡改结果包括已篡改结果和未篡改结果；
D、对各个图像样本进行块分割，将各个图像样本分成N*M(例如，32*32)的图像小碎片，N和M为正整数，并从各个图像样本的各个图像小碎片中提取出初始篡改检测特征，以提取出各个图像样本对应的初始篡改检测特征；
E、将所有图像样本按照X：Y(例如，8:2)的比例分成第一数据集和第二数据集，第一数据集中的图像样本数量大于第二数据集中的图像样本数量，第一数据集作为训练集，第二数据集作为测试集，其中，X大于0，Y大于0；
F、利用第一数据集中的各个图像样本对应的初始篡改检测特征进行编码器的训练，利用第二数据集中的各个图像样本对应的初始篡改检测特征对训练的编码器进行准确率验证；
G、若训练的编码器的准确率大于或等于预设准确率(例如，95％)，则结束训练，或者，若训练的编码器的准确率小于预设准确率，则增加图像样本的数量，并重复执行上述步骤D、E和F，直至训练的编码器的准确率大于或等于预设准确率。
本实施例通过对待检测图像进行块分割成若干图像小碎片，从各个图像小碎片中提取初始篡改检测特征，并利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，根据该复杂 篡改特征确定所述待检测图像的篡改检测结果。由于是将待检测图像进行分割后，基于各个图像小碎片的初始篡改检测特征以及利用编码器对该初始篡改检测特征编码生成的复杂篡改特征来对待检测图像的篡改进行检测，并不依赖JPEG压缩痕迹原理，因此，能进行篡改检测的图像类型并不仅限于JPEG，而且，通过分割及编码的方式进行检测，能准确地对多种类型的图像篡改方式进行检测，从而实现对不同类型、格式的图像篡改进行准确的检测。
进一步地，如图3所示，上述步骤S10可以包括：
步骤S101，将所述待检测图像从第一颜色空间(例如，RGB颜色空间)转换到第二颜色空间(例如，YCrCb颜色空间)；
步骤S102，对所述待检测图像进行块分割，将所述待检测图像分成N*M(例如，32*32)的图像小碎片，N和M为正整数；
步骤S103，在每个图像小碎片上的每个第二颜色空间分量上应用预设的小波函数如多个级别(例如，3个级别)的二维多贝西小波(Daubechies Wavelet)分解，以得到多个(例如，30个)对应的像素系数映射(Coefficient Map)，对各个所述像素系数映射计算汇总统计系数(例如，每个像素系数映射对应的像素标准差、像素均值和/或像素总和)，并应用小波函数如多贝西正交(Daubechies Orthogonal wavelets)D2-D5在每个图像小碎片上获得多个汇总统计系数，各个所述汇总统计系数即是对应的图像小碎片的初始篡改检测特征。
本实施例中，将所述待检测图像进行块分割成若干图像小碎片进行特征提取，能更细致的提取所述待检测图像的每一细节特征，而且，利用多贝西小波、多贝西正交等方式来获取每一图像小碎片的汇总统计系数作为初始篡改检测特征，能使得提取的初始篡改检测特征更加精确。
进一步地，在其他实施例中，所述预先确定的编码器为栈化自动编码器(Stacked Auto-Encoder，SAE)，参照图4，图4为本发明图像篡改检测方法一实施例中栈化自动编码器的结构示意图。该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入。该栈化自动编码器还包括一个神经网络多层感知器(Multi-Layer perceptron，MLP)，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。例如，假设W(l)，b(l)表示第l-th层自动编码器神经元的权值矩阵和偏置向量，l-th层自动编码器的输入为Z(l)，输出为ae(l)，则：
e(l)＝f(Z(l))，Z(l+1)＝W(l)ae(l)+b(l)，
其中，f(.)是神经元的激活函数，在本实施例中使用S型函数，其表达式如下：
Figure PCTCN2017091312-appb-000001
SAE的解码过程是编码过程的逆运算，相关输入和输出表达式如下：
ae(l)＝f(Z(n+l))，Z(n+l+1)＝W(n-l)ae(n+l)+b(n-l)，
其中n表示编码层总的层数，从以上公式可以看出SAE中编码层的层数与解码层层数一致，但测试时只使用编码层。因此，在使用SAE提取输入中的特征时，最后的特征表示即是最后一层的激活值向量Y，其表达式如下：
Y＝ae(n)。
本实施例中，该栈化自动编码器输入层接收初始篡改检测特征，然后将接收的初始篡改检测特征传入中间隐含编码层进行特征降维过滤，再经过解码层的解码重构出原来的输入层，通过计算重构的输入与原输入之间的差异，从而指导SAE的编码和解码层的训练，最终确保SAE编码层的最后输出是原始输入特征的高级复杂篡改特征，使用该高级复杂篡改特征有利于提升SAE提取的特征的判定分类能力，进而提高图像篡改检测的准确性。
进一步地，在其他实施例中，上述步骤S20可以包括：
利用预先确定的编码器对各个图像小碎片的初始篡改检测特征进行编码，以生成各个图像小碎片的复杂篡改特征；
根据各个图像小碎片的复杂篡改特征确定出各个图像小碎片对应的第一篡改检测结果，所述第一篡改检测结果包括已篡改和未篡改；
若有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改。
本实施例中，在利用预先确定的编码器如栈化自动编码器对各个图像小碎片的初始篡改检测特征进行编码生成对应的复杂篡改特征后，即可根据各个图像小碎片的复杂篡改特征来对各个图像小碎片是否篡改进行检测。若存在有一个或多个图像小碎片对应的第一篡改检测结果为已篡改，则说明待检测图像有被篡改的部分，则确定所述待检测图像对应的篡改检测结果为已篡改。若所有图像小碎片对应的第一篡改检测结果均为未篡改，则说明待检测图像没有被篡改的部分， 则确定所述待检测图像对应的篡改检测结果为未篡改。
进一步地，在其他实施例中，该栈化自动编码器还包括一个用于根据相邻区域的场景信息确定出所述待检测图像对应的篡改检测结果的相邻区域感知层，上述步骤S20还可以包括：
根据预设的算法对各个图像小碎片的临近场景信息进行计算，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果，所述第二篡改检测结果包括已篡改和未篡改；
若有图像小碎片对应的第二篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改；
若所有图像小碎片对应的第二篡改检测结果均为未篡改，且有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改。
本实施例中，在提取各个图像小碎片的复杂篡改特征的基础上，还利用相邻区域感知层来计算获取各个图像小碎片的临近场景信息，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果。在综合考虑第一篡改检测结果、第二篡改检测结果的基础上来进行所述待检测图像的篡改检测。如若所有图像小碎片对应的第二篡改检测结果均为未篡改，而有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改，防止因第一篡改检测结果误判等原因导致所述待检测图像出现篡改检测错误的情况发生，进一步地提高了图像篡改检测的准确性及稳定性。
在一种实施方式中，所述预设的算法包括：
对一个图像小碎片P，假设它的临近场景信息为N(p)，则N(p)的表达式如下：
Figure PCTCN2017091312-appb-000002
公式中的
Figure PCTCN2017091312-appb-000003
表示该图像小碎片P的第i个相邻图像小碎片的复杂篡改特征，K表示该图像小碎片P总共有K个相邻的图像小碎片。计算出该图像小碎片P对应的所有相邻的图像小碎片的分类结果的平均值，所述平均值的计算公式为：
Figure PCTCN2017091312-appb-000004
其中，
Figure PCTCN2017091312-appb-000005
表示该图像小碎片P的第i个相邻图像小碎片的MLP分类结果，K可以被设定为8(即：每个图像小碎片的8个可能方向均会被用作最终篡改判定分类)；若该图像小碎片P对应的分类结果的平均值大于等于预设阈值(例如，0.5)，则确定该图像小碎 片P对应的第二篡改检测结果为已篡改结果，或者，若该图像小碎片P对应的分类结果的平均值小于预设阈值，则确定当前图像小碎片P对应的第二篡改检测结果为未篡改结果。
本发明进一步提供一种图像篡改检测系统。
参照图5，图5为本发明图像篡改检测系统一实施例的功能模块示意图。
在一实施例中，该图像篡改检测系统包括：
提取模块01，用于对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征。
本实施例中，对如JPEG、PNG、GIF等各种类型的待检测图像进行块分割，将所述待检测图像分成N*M(例如，32*32)的图像小碎片，N和M为正整数，并从各个图像小碎片中提取出初始篡改检测特征，该初始篡改检测特征可包括颜色特征、像素特征等，在此不做限定。其中，对待检测图像进行块分割时，可以将所述待检测图像等分成各个相同大小的图像小碎片，也可以将所述待检测图像按一定比例或随机分成不同大小的图像小碎片，在此不做限定。图像小碎片既可以是形状规则的正方形、长方形等，也可以是形状不规则的碎片，在此不做限定。
检测模块02，用于利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
提取出各个图像小碎片中的初始篡改检测特征之后，利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征最终进行检测确定所述待检测图像是已篡改或未篡改的状态。其中，该预先确定的编码器可以是相对固定的编码器，也可以是能进行学习、训练等深度学习的编码器，例如可以是单自动编码器、栈式自动编码器等，在此不做限定。
在一种实施方式中，该预先确定的编码器是能进行学习、训练等深度学习的编码器如栈式自动编码器等，该预先确定的编码器的训练过程如下：
C、获取预设数量(例如，10万个)的图像样本，并对各个图像样本的篡改结果进行标识，所述篡改结果包括已篡改结果和未篡改结果；
D、对各个图像样本进行块分割，将各个图像样本分成N*M(例如，32*32)的图像小碎片，N和M为正整数，并从各个图像样本的 各个图像小碎片中提取出初始篡改检测特征，以提取出各个图像样本对应的初始篡改检测特征；
E、将所有图像样本按照X：Y(例如，8:2)的比例分成第一数据集和第二数据集，第一数据集中的图像样本数量大于第二数据集中的图像样本数量，第一数据集作为训练集，第二数据集作为测试集，其中，X大于0，Y大于0；
F、利用第一数据集中的各个图像样本对应的初始篡改检测特征进行编码器的训练，利用第二数据集中的各个图像样本对应的初始篡改检测特征对训练的编码器进行准确率验证；
G、若训练的编码器的准确率大于或等于预设准确率(例如，95％)，则结束训练，或者，若训练的编码器的准确率小于预设准确率，则增加图像样本的数量，并重复执行上述步骤D、E和F，直至训练的编码器的准确率大于或等于预设准确率。
本实施例通过对待检测图像进行块分割成若干图像小碎片，从各个图像小碎片中提取初始篡改检测特征，并利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，根据该复杂篡改特征确定所述待检测图像的篡改检测结果。由于是将待检测图像进行分割后，基于各个图像小碎片的初始篡改检测特征以及利用编码器对该初始篡改检测特征编码生成的复杂篡改特征来对待检测图像的篡改进行检测，并不依赖JPEG压缩痕迹原理，因此，能进行篡改检测的图像类型并不仅限于JPEG，而且，通过分割及编码的方式进行检测，能准确地对多种类型的图像篡改方式进行检测，从而实现对不同类型、格式的图像篡改进行准确的检测。
进一步地，如图6所示，上述提取模块01可以包括：
转换单元011，用于将所述待检测图像从第一颜色空间(例如，RGB颜色空间)转换到第二颜色空间(例如，YCrCb颜色空间)；
分割单元012，用于对所述待检测图像进行块分割，将所述待检测图像分成N*M(例如，32*32)的图像小碎片，N和M为正整数；
获取单元013，用于在每个图像小碎片上的每个第二颜色空间分量上应用预设的小波函数如多个级别(例如，3个级别)的二维多贝西小波(Daubechies Wavelet)分解，以得到多个(例如，30个)对应的像素系数映射(Coefficient Map)，对各个所述像素系数映射计算汇总统计系数(例如，每个像素系数映射对应的像素标准差、像素均值和/或像素总和)，并应用小波函数如多贝西正交(Daubechies Orthogonal wavelets)D2-D5在每个图像小碎片上获得多个汇总统计系数，各个所述汇总统计系数即是对应的图像小碎片的初始篡改检测 特征。
本实施例中，将所述待检测图像进行块分割成若干图像小碎片进行特征提取，能更细致的提取所述待检测图像的每一细节特征，而且，利用多贝西小波、多贝西正交等方式来获取每一图像小碎片的汇总统计系数作为初始篡改检测特征，能使得提取的初始篡改检测特征更加精确。
进一步地，在其他实施例中，所述预先确定的编码器为栈化自动编码器(Stacked Auto-Encoder，SAE)，参照图4，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入。该栈化自动编码器还包括一个神经网络多层感知器(Multi-Layer perceptron，MLP)，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。例如，假设W(l)，b(l)表示第l-th层自动编码器神经元的权值矩阵和偏置向量，l-th层自动编码器的输入为Z(l)，输出为ae(l)，则：
e(l)＝f(Z(l))，Z(l+1)＝W(l)ae(l)+b(l)，
其中，f(.)是神经元的激活函数，在本实施例中使用S型函数，其表达式如下：
Figure PCTCN2017091312-appb-000006
SAE的解码过程是编码过程的逆运算，相关输入和输出表达式如下：
ae(l)＝f(Z(n+l))，Z(n+l+1)＝W(n-l)ae(n+l)+b(n-l)，
其中n表示编码层总的层数，从以上公式可以看出SAE中编码层的层数与解码层层数一致，但测试时只使用编码层。因此，在使用SAE提取输入中的特征时，最后的特征表示即是最后一层的激活值向量Y，其表达式如下：
Y＝ae(n)。
本实施例中，该栈化自动编码器输入层接收初始篡改检测特征，然后将接收的初始篡改检测特征传入中间隐含编码层进行特征降维过滤，再经过解码层的解码重构出原来的输入层，通过计算重构的输入与原输入之间的差异，从而指导SAE的编码和解码层的训练，最终确保SAE编码层的最后输出是原始输入特征的高级复杂篡改特征，使用该高级复杂篡改特征有利于提升SAE提取的特征的判定分类能 力，进而提高图像篡改检测的准确性。
进一步地，在其他实施例中，上述检测模块02还可以用于：
利用预先确定的编码器对各个图像小碎片的初始篡改检测特征进行编码，以生成各个图像小碎片的复杂篡改特征；
根据各个图像小碎片的复杂篡改特征确定出各个图像小碎片对应的第一篡改检测结果，所述第一篡改检测结果包括已篡改和未篡改；
若有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改。
本实施例中，在利用预先确定的编码器如栈化自动编码器对各个图像小碎片的初始篡改检测特征进行编码生成对应的复杂篡改特征后，即可根据各个图像小碎片的复杂篡改特征来对各个图像小碎片是否篡改进行检测。若存在有一个或多个图像小碎片对应的第一篡改检测结果为已篡改，则说明待检测图像有被篡改的部分，则确定所述待检测图像对应的篡改检测结果为已篡改。若所有图像小碎片对应的第一篡改检测结果均为未篡改，则说明待检测图像没有被篡改的部分，则确定所述待检测图像对应的篡改检测结果为未篡改。
进一步地，在其他实施例中，该栈化自动编码器还包括一个用于根据相邻区域的场景信息确定出所述待检测图像对应的篡改检测结果的相邻区域感知层，上述检测模块02还可以用于
根据预设的算法对各个图像小碎片的临近场景信息进行计算，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果，所述第二篡改检测结果包括已篡改和未篡改；
若有图像小碎片对应的第二篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改；
若所有图像小碎片对应的第二篡改检测结果均为未篡改，且有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改。
本实施例中，在提取各个图像小碎片的复杂篡改特征的基础上，还利用相邻区域感知层来计算获取各个图像小碎片的临近场景信息，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果。在综合考虑第一篡改检测结果、第二篡改检测结果的基础上来进行所述待检测图像的篡改检测。如若所有图像小碎片对应的第二篡改检测结果均为未篡改，而有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改，防止因第一篡改检测结果误判等原因导致所述待检测图像 出现篡改检测错误的情况发生，进一步地提高了图像篡改检测的准确性及稳定性。
在一种实施方式中，所述预设的算法包括：
对一个图像小碎片P，假设它的临近场景信息为N(p)，则N(p)的表达式如下：
Figure PCTCN2017091312-appb-000007
公式中的
Figure PCTCN2017091312-appb-000008
表示该图像小碎片P的第i个相邻图像小碎片的复杂篡改特征，K表示该图像小碎片P总共有K个相邻的图像小碎片。计算出该图像小碎片P对应的所有相邻的图像小碎片的分类结果的平均值，所述平均值的计算公式为：
Figure PCTCN2017091312-appb-000009
其中，
Figure PCTCN2017091312-appb-000010
表示该图像小碎片P的第i个相邻图像小碎片的MLP分类结果，K可以被设定为8(即：每个图像小碎片的8个可能方向均会被用作最终篡改判定分类)；若该图像小碎片P对应的分类结果的平均值大于等于预设阈值(例如，0.5)，则确定该图像小碎片P对应的第二篡改检测结果为已篡改结果，或者，若该图像小碎片P对应的分类结果的平均值小于预设阈值，则确定当前图像小碎片P对应的第二篡改检测结果为未篡改结果。
在硬件实现上，以上提取模块01、检测模块02等可以以硬件形式内嵌于或独立于电子装置中，也可以以软件形式存储于电子装置的存储设备中，以便于处理设备调用执行以上各个模块对应的操作。该处理设备可以为中央处理单元(CPU)、微处理器、单片机等。
此外，本发明还提供一种计算机可读存储介质，所述计算机可读存储介质存储有图像篡改检测系统，所述图像篡改检测系统可被至少一个处理器执行，以使所述至少一个处理器执行如上述实施例中的图像篡改检测方法的步骤，该图像篡改检测方法的步骤S10、S20、S30等具体实施过程如上文所述，在此不再赘述。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存 在另外的相同要素。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件来实现，但很多情况下前者是更佳的实施方式。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本发明各个实施例所述的方法。
以上参照附图说明了本发明的优选实施例，并非因此局限本发明的权利范围。上述本发明实施例序号仅仅为了描述，不代表实施例的优劣。另外，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本领域技术人员不脱离本发明的范围和实质，可以有多种变型方案实现本发明，比如作为一个实施例的特征可用于另一实施例而得到又一实施例。凡在运用本发明的技术构思之内所作的任何修改、等同替换和改进，均应在本发明的权利范围之内。

Claims (20)

  1. 一种图像篡改检测方法，其特征在于，所述方法包括以下步骤：
    A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
    B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
  2. 如权利要求1所述的图像篡改检测方法，其特征在于，所述步骤A包括：
    将所述待检测图像从第一颜色空间转换到第二颜色空间；
    对所述待检测图像进行块分割，将所述待检测图像分成N*M的图像小碎片，N和M为正整数；
    在每个图像小碎片的每个第二颜色空间分量上应用预设的小波函数分解，以得到多个对应的像素系数映射，对各个所述像素系数映射计算汇总统计系数，在每个图像小碎片上获得多个汇总统计系数，将各个所述汇总统计系数作为对应的图像小碎片的初始篡改检测特征。
  3. 如权利要求1所述的图像篡改检测方法，其特征在于，所述预先确定的编码器为栈化自动编码器，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入，该栈化自动编码器还包括一个神经网络多层感知器，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。
  4. 如权利要求2所述的图像篡改检测方法，其特征在于，所述预先确定的编码器为栈化自动编码器，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入，该栈化自动编码器还包括一个神经网络多层感知器，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。
  5. 如权利要求3所述的图像篡改检测方法，其特征在于，所述步骤B包括：
    利用预先确定的编码器对各个图像小碎片的初始篡改检测特征进行编码，以生成各个图像小碎片的复杂篡改特征；
    根据各个图像小碎片的复杂篡改特征确定出各个图像小碎片对应的第一篡改检测结果，所述第一篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改。
  6. 如权利要求5所述的图像篡改检测方法，其特征在于，该栈化自动编码器还包括一个用于根据相邻区域的场景信息确定出所述待检测图像对应的篡改检测结果的相邻区域感知层，所述步骤B还包括：
    根据预设的算法对各个图像小碎片的临近场景信息进行计算，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果，所述第二篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第二篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改；
    若所有图像小碎片对应的第二篡改检测结果均为未篡改，且有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改。
  7. 如权利要求1所述的图像篡改检测方法，其特征在于，所述预先确定的编码器的训练过程如下：
    C、获取预设数量的图像样本，并对各个图像样本的篡改结果进行标识，所述篡改结果包括已篡改和未篡改；
    D、对各个图像样本进行块分割，将各个图像样本分成N*M的图像小碎片，N和M为正整数，并从各个图像样本的各个图像小碎片中提取出初始篡改检测特征，以提取出各个图像样本对应的初始篡改检测特征；
    E、将所有图像样本按照X：Y的比例分成第一数据集和第二数据集，第一数据集中的图像样本数量大于第二数据集中的图像样本数量，第一数据集作为训练集，第二数据集作为测试集，其中，X大于0，Y大于0；
    F、利用第一数据集中的各个图像样本对应的初始篡改检测特征进行编码器的训练，利用第二数据集中的各个图像样本对应的初始篡改检测特征对训练的编码器进行准确率验证；
    G、若训练的编码器的准确率大于或等于预设准确率，则结束训练，或者，若训练的编码器的准确率小于预设准确率，则增加图像样本的数量，并重复执行上述步骤D、E和F，直至训练的编码器的准确率大于或等于预设准确率。
  8. 一种图像篡改检测系统，其特征在于，所述图像篡改检测系统包括：
    提取模块，用于对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
    检测模块，用于利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
  9. 如权利要求8所述的图像篡改检测系统，其特征在于，所述提取模块包括：
    转换单元，用于将所述待检测图像从第一颜色空间转换到第二颜色空间；
    分割单元，用于对所述待检测图像进行块分割，将所述待检测图像分成N*M的图像小碎片，N和M为正整数；
    获取单元，用于在每个图像小碎片的每个第二颜色空间分量上应用预设的小波函数分解，以得到多个对应的像素系数映射，对各个所述像素系数映射计算汇总统计系数，在每个图像小碎片上获得多个汇总统计系数，将各个所述汇总统计系数作为对应的图像小碎片的初始篡改检测特征。
  10. 如权利要求8所述的图像篡改检测系统，其特征在于，所述预先确定的编码器为栈化自动编码器，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入，该栈化自动编码器还包括一个神经网络多层感知器，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。
  11. 如权利要求9所述的图像篡改检测系统，其特征在于，所述预先确定的编码器为栈化自动编码器，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入，该栈化自动编码器还包括一个神经网络多层感知器，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。
  12. 如权利要求10所述的图像篡改检测系统，其特征在于，所述检测模块还用于：
    利用预先确定的编码器对各个图像小碎片的初始篡改检测特征进行编码，以生成各个图像小碎片的复杂篡改特征；
    根据各个图像小碎片的复杂篡改特征确定出各个图像小碎片对应的第一篡改检测结果，所述第一篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改。
  13. 如权利要求12所述的图像篡改检测系统，其特征在于，该栈化自动编码器还包括一个用于根据相邻区域的场景信息确定出所述待检测图像对应的篡改检测结果的相邻区域感知层，所述检测模块还用于：
    根据预设的算法对各个图像小碎片的临近场景信息进行计算，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果，所述第二篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第二篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改；
    若所有图像小碎片对应的第二篡改检测结果均为未篡改，且有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改。
  14. 如权利要求8所述的图像篡改检测系统，其特征在于，所述预先确定的编码器的训练过程如下：
    C、获取预设数量的图像样本，并对各个图像样本的篡改结果进行标识，所述篡改结果包括已篡改和未篡改；
    D、对各个图像样本进行块分割，将各个图像样本分成N*M的图像小碎片，N和M为正整数，并从各个图像样本的各个图像小碎片中提取出初始篡改检测特征，以提取出各个图像样本对应的初始篡改检测特征；
    E、将所有图像样本按照X：Y的比例分成第一数据集和第二数据集，第一数据集中的图像样本数量大于第二数据集中的图像样本数量，第一数据集作为训练集，第二数据集作为测试集，其中，X大于0，Y大于0；
    F、利用第一数据集中的各个图像样本对应的初始篡改检测特征进行编码器的训练，利用第二数据集中的各个图像样本对应的初始篡改检测特征对训练的编码器进行准确率验证；
    G、若训练的编码器的准确率大于或等于预设准确率，则结束训练，或者，若训练的编码器的准确率小于预设准确率，则增加图像样 本的数量，并重复执行上述步骤D、E和F，直至训练的编码器的准确率大于或等于预设准确率。
  15. 一种电子装置，其特征在于，包括处理设备及存储设备；该存储设备中存储有图像篡改检测程序，其包括至少一个计算机可读指令，该至少一个计算机可读指令可被所述处理设备执行，以实现以下操作：
    A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
    B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
  16. 如权利要求15所述的电子装置，其特征在于，所述至少一个计算机可读指令被所述处理设备执行，在执行所述步骤A时包括：
    将所述待检测图像从第一颜色空间转换到第二颜色空间；
    对所述待检测图像进行块分割，将所述待检测图像分成N*M的图像小碎片，N和M为正整数；
    在每个图像小碎片的每个第二颜色空间分量上应用预设的小波函数分解，以得到多个对应的像素系数映射，对各个所述像素系数映射计算汇总统计系数，在每个图像小碎片上获得多个汇总统计系数，将各个所述汇总统计系数作为对应的图像小碎片的初始篡改检测特征。
  17. 如权利要求15所述的电子装置，其特征在于，所述预先确定的编码器为栈化自动编码器，该栈化自动编码器由多层基本的自动编码器层叠而成，每一层自动编码器的输出是下一层自动编码器的输入，该栈化自动编码器还包括一个神经网络多层感知器，该神经网络多层感知器与最后一层的自动编码器对接，用于根据生成的复杂篡改特征确定出所述待检测图像对应的篡改检测结果。
  18. 如权利要求17所述的电子装置，其特征在于，所述至少一个计算机可读指令被所述处理设备执行，在执行所述步骤B时包括：
    利用预先确定的编码器对各个图像小碎片的初始篡改检测特征进行编码，以生成各个图像小碎片的复杂篡改特征；
    根据各个图像小碎片的复杂篡改特征确定出各个图像小碎片对应的第一篡改检测结果，所述第一篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改。
  19. 如权利要求18所述的电子装置，其特征在于，该栈化自动编码器还包括一个用于根据相邻区域的场景信息确定出所述待检测图像对应的篡改检测结果的相邻区域感知层，所述至少一个计算机可读指令被所述处理设备执行，在执行所述步骤B时包括：
    根据预设的算法对各个图像小碎片的临近场景信息进行计算，并根据各个图像小碎片的临近场景信息确定出各个图像小碎片对应的第二篡改检测结果，所述第二篡改检测结果包括已篡改和未篡改；
    若有图像小碎片对应的第二篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为已篡改；
    若所有图像小碎片对应的第二篡改检测结果均为未篡改，且有图像小碎片对应的第一篡改检测结果为已篡改，则确定所述待检测图像对应的篡改检测结果为未篡改。
  20. 一种计算机可读存储介质，其上存储有至少一个可被处理设备执行以实现以下操作的计算机可读指令：
    A、对待检测图像进行块分割，将所述待检测图像分成若干图像小碎片，并从各个图像小碎片中提取出初始篡改检测特征；
    B、利用预先确定的编码器对提取的初始篡改检测特征进行编码以生成复杂篡改特征，并根据生成的复杂篡改特征确定所述待检测图像对应的篡改检测结果，所述篡改检测结果包括已篡改和未篡改。
PCT/CN2017/091312 2016-12-30 2017-06-30 图像篡改检测方法、系统、电子装置及存储介质 WO2018120724A1 (zh)

Priority Applications (6)

Application Number Priority Date Filing Date Title
AU2017389535A AU2017389535B2 (en) 2016-12-30 2017-06-30 Image tampering detection method and system, electronic apparatus and storage medium
SG11201808824YA SG11201808824YA (en) 2016-12-30 2017-06-30 Method and system of detecting image tampering, electronic device and storage medium
EP17882272.2A EP3396625A4 (en) 2016-12-30 2017-06-30 METHOD AND SYSTEM FOR ALTERATION DETECTION OF IMAGE, ELECTRONIC APPARATUS AND STORAGE MEDIUM
KR1020187017248A KR102168397B1 (ko) 2016-12-30 2017-06-30 이미지 탬퍼 검출 방법, 시스템, 전자장치 및 저장매체
JP2018528057A JP2019503530A (ja) 2016-12-30 2017-06-30 画像改ざん検出方法、システム、電子装置及び記憶媒体
US16/084,227 US10692218B2 (en) 2016-12-30 2017-06-30 Method and system of detecting image tampering, electronic device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611265608.2 2016-12-30
CN201611265608.2A CN106846303A (zh) 2016-12-30 2016-12-30 图像篡改检测方法及装置

Publications (1)

Publication Number Publication Date
WO2018120724A1 true WO2018120724A1 (zh) 2018-07-05

Family

ID=59117401

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/091312 WO2018120724A1 (zh) 2016-12-30 2017-06-30 图像篡改检测方法、系统、电子装置及存储介质

Country Status (9)

Country Link
US (1) US10692218B2 (zh)
EP (1) EP3396625A4 (zh)
JP (1) JP2019503530A (zh)
KR (1) KR102168397B1 (zh)
CN (1) CN106846303A (zh)
AU (1) AU2017389535B2 (zh)
SG (1) SG11201808824YA (zh)
TW (1) TWI665639B (zh)
WO (1) WO2018120724A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112465764A (zh) * 2020-11-24 2021-03-09 泰康保险集团股份有限公司 一种图像篡改检测方法和装置
CN112861960A (zh) * 2021-02-03 2021-05-28 湖南大学 一种图像篡改检测方法、系统及存储介质
CN114612798A (zh) * 2022-03-09 2022-06-10 云南大学 基于Flow模型的卫星图像篡改检测方法
CN115965581A (zh) * 2022-11-22 2023-04-14 澳门科技大学 一种复制-粘贴篡改图像检测方法、系统及设备
CN116342601A (zh) * 2023-05-30 2023-06-27 山东省人工智能研究院 基于边缘引导和多层级搜索的图像篡改检测方法

Families Citing this family (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106846303A (zh) * 2016-12-30 2017-06-13 平安科技（深圳）有限公司 图像篡改检测方法及装置
CN107464237A (zh) * 2017-08-04 2017-12-12 平安科技（深圳）有限公司 图像篡改检测方法、电子装置及可读存储介质
CN107657259A (zh) * 2017-09-30 2018-02-02 平安科技（深圳）有限公司 图像篡改检测方法、电子装置及可读存储介质
CN108230227B (zh) * 2018-02-06 2021-09-03 中国银行股份有限公司 一种图像篡改的识别方法、装置及电子设备
CN108564035B (zh) * 2018-04-13 2020-09-25 杭州睿琪软件有限公司 识别单据上记载的信息的方法及系统
CN111709883B (zh) 2019-03-01 2023-06-13 阿里巴巴集团控股有限公司 一种图像检测方法、装置及设备
CN111753595A (zh) * 2019-03-29 2020-10-09 北京市商汤科技开发有限公司 活体检测方法和装置、设备和存储介质
CN110009621B (zh) * 2019-04-02 2023-11-07 广东工业大学 一种篡改视频检测方法、装置、设备及可读存储介质
CN110378254B (zh) * 2019-07-03 2022-04-19 中科软科技股份有限公司 车损图像修改痕迹的识别方法、系统、电子设备及存储介质
CN110609794B (zh) * 2019-09-12 2023-04-28 中国联合网络通信集团有限公司 页面检测方法及装置
CN110895811B (zh) * 2019-11-05 2023-05-09 泰康保险集团股份有限公司 一种图像篡改检测方法和装置
CN110942456B (zh) * 2019-11-25 2024-01-23 深圳前海微众银行股份有限公司 篡改图像检测方法、装置、设备及存储介质
CN111275055B (zh) * 2020-01-21 2023-06-06 北京市商汤科技开发有限公司 网络训练方法及装置、图像处理方法及装置
CN111325265B (zh) * 2020-02-17 2023-09-01 中国银联股份有限公司 一种针对篡改图像的检测方法及装置
CN111260645B (zh) * 2020-02-20 2023-10-13 中国科学院自动化研究所 基于分块分类深度学习的篡改图像检测方法及系统
CN111553916B (zh) * 2020-05-09 2023-11-14 中科计算技术创新研究院 基于多种特征和卷积神经网络的图像篡改区域检测方法
CN111950037A (zh) * 2020-08-25 2020-11-17 北京天融信网络安全技术有限公司 检测方法、装置、电子设备及存储介质
CN112508039B (zh) * 2020-12-08 2024-04-02 中国银联股份有限公司 一种图像检测方法及装置
CN112949431B (zh) * 2021-02-08 2024-06-25 证通股份有限公司 视频篡改检测方法和系统、存储介质
CN112950564B (zh) * 2021-02-23 2022-04-01 北京三快在线科技有限公司 一种图像检测方法、装置、存储介质及电子设备
CN112800727B (zh) * 2021-04-14 2021-07-20 北京三维天地科技股份有限公司 给pdf文件加批注的方法及应用系统
CN113012152B (zh) * 2021-04-27 2023-04-14 深圳大学 一种图像篡改链检测方法、装置及电子设备
CN113780062A (zh) * 2021-07-26 2021-12-10 岚图汽车科技有限公司 一种基于情感识别的车载智能交互方法、存储介质及芯片
CN114677670B (zh) * 2022-03-30 2024-04-26 康旭科技有限公司 一种身份证篡改自动识别与定位的方法
CN114612476B (zh) * 2022-05-13 2022-07-22 南京信息工程大学 一种基于全分辨率混合注意力机制的图像篡改检测方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101655972A (zh) * 2009-09-17 2010-02-24 上海交通大学 基于小波域的图像拼接盲检测方法
CN102063627A (zh) * 2010-12-31 2011-05-18 宁波大学 基于多小波变换的自然图像和计算机生成图像的识别方法
CN102184537A (zh) * 2011-04-22 2011-09-14 西安理工大学 基于小波变换和主成分分析的图像区域篡改检测方法
CN103345758A (zh) * 2013-07-25 2013-10-09 南京邮电大学 基于dct统计特征的jpeg图像区域复制篡改盲检测方法
US20150339543A1 (en) * 2014-05-22 2015-11-26 Xerox Corporation Method and apparatus for classifying machine printed text and handwritten text
CN105426912A (zh) * 2015-11-12 2016-03-23 河南师范大学 一种置换混叠图像的盲分离方法
CN106846303A (zh) * 2016-12-30 2017-06-13 平安科技（深圳）有限公司 图像篡改检测方法及装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6285775B1 (en) * 1998-10-01 2001-09-04 The Trustees Of The University Of Princeton Watermarking scheme for image authentication
US7778461B2 (en) * 2006-05-05 2010-08-17 New Jersey Institute Of Technology System and/or method for image tamper detection
ATE521054T1 (de) * 2006-12-20 2011-09-15 Axis Ab Verfahen und einrichtung zur erkennung von sabotage an einer überwachungskamera
US8023747B2 (en) * 2007-02-09 2011-09-20 New Jersey Institute Of Technology Method and apparatus for a natural image model based approach to image/splicing/tampering detection
US7792377B2 (en) * 2007-04-25 2010-09-07 Huper Laboratories Co., Ltd. Method of image authentication and restoration
CN101901470A (zh) * 2010-02-10 2010-12-01 桂林电子科技大学 基于能量域半脆弱水印的图像篡改检测及恢复方法
CN103544703B (zh) * 2013-10-19 2016-12-07 上海理工大学 数字图像拼接检测方法
TWI534651B (zh) * 2014-01-21 2016-05-21 Nat Penghu University Of Science And Technology Image forgery detection method based on transparent layer mask
TW201612854A (en) * 2014-09-23 2016-04-01 Nat Penghu University Of Science And Technology Image component feature based synthetic image forgery detection method
CN109903302B (zh) * 2015-06-25 2022-11-04 北京影谱科技股份有限公司 一种用于拼接图像的篡改检测方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101655972A (zh) * 2009-09-17 2010-02-24 上海交通大学 基于小波域的图像拼接盲检测方法
CN102063627A (zh) * 2010-12-31 2011-05-18 宁波大学 基于多小波变换的自然图像和计算机生成图像的识别方法
CN102184537A (zh) * 2011-04-22 2011-09-14 西安理工大学 基于小波变换和主成分分析的图像区域篡改检测方法
CN103345758A (zh) * 2013-07-25 2013-10-09 南京邮电大学 基于dct统计特征的jpeg图像区域复制篡改盲检测方法
US20150339543A1 (en) * 2014-05-22 2015-11-26 Xerox Corporation Method and apparatus for classifying machine printed text and handwritten text
CN105426912A (zh) * 2015-11-12 2016-03-23 河南师范大学 一种置换混叠图像的盲分离方法
CN106846303A (zh) * 2016-12-30 2017-06-13 平安科技（深圳）有限公司 图像篡改检测方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3396625A4 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112465764A (zh) * 2020-11-24 2021-03-09 泰康保险集团股份有限公司 一种图像篡改检测方法和装置
CN112861960A (zh) * 2021-02-03 2021-05-28 湖南大学 一种图像篡改检测方法、系统及存储介质
CN112861960B (zh) * 2021-02-03 2022-09-02 湖南大学 一种图像篡改检测方法、系统及存储介质
CN114612798A (zh) * 2022-03-09 2022-06-10 云南大学 基于Flow模型的卫星图像篡改检测方法
CN114612798B (zh) * 2022-03-09 2024-05-14 云南大学 基于Flow模型的卫星图像篡改检测方法
CN115965581A (zh) * 2022-11-22 2023-04-14 澳门科技大学 一种复制-粘贴篡改图像检测方法、系统及设备
CN115965581B (zh) * 2022-11-22 2023-09-12 澳门科技大学 一种复制-粘贴篡改图像检测方法、系统及设备
CN116342601A (zh) * 2023-05-30 2023-06-27 山东省人工智能研究院 基于边缘引导和多层级搜索的图像篡改检测方法
CN116342601B (zh) * 2023-05-30 2023-07-21 山东省人工智能研究院 基于边缘引导和多层级搜索的图像篡改检测方法

Also Published As

Publication number Publication date
AU2017389535A1 (en) 2018-10-04
KR20190019040A (ko) 2019-02-26
US10692218B2 (en) 2020-06-23
US20190156486A1 (en) 2019-05-23
CN106846303A (zh) 2017-06-13
KR102168397B1 (ko) 2020-10-22
EP3396625A4 (en) 2019-08-07
EP3396625A1 (en) 2018-10-31
TW201824179A (zh) 2018-07-01
TWI665639B (zh) 2019-07-11
AU2017389535B2 (en) 2019-10-10
JP2019503530A (ja) 2019-02-07
SG11201808824YA (en) 2018-11-29

Similar Documents

Publication Publication Date Title
WO2018120724A1 (zh) 图像篡改检测方法、系统、电子装置及存储介质
CN108038474B (zh) 人脸检测方法、卷积神经网络参数的训练方法、装置及介质
WO2019061661A1 (zh) 图像篡改检测方法、电子装置及可读存储介质
CN107622489B (zh) 一种图像篡改检测方法及装置
CN108345827B (zh) 识别文档方向的方法、系统和神经网络
Lubenko et al. Steganalysis with mismatched covers: Do simple classifiers help?
AU2017209231A1 (en) Method, system, device and readable storage medium for realizing insurance claim fraud prevention based on consistency between multiple images
CN110765860A (zh) 摔倒判定方法、装置、计算机设备及存储介质
RU2018145499A (ru) Автоматизация проверки достоверности изображения
WO2015170461A1 (ja) 画像処理装置、画像処理方法およびコンピュータ可読記録媒体
CN111178367B (zh) 适应多物件尺寸的特征决定装置及方法
CN110618854B (zh) 基于深度学习与内存镜像分析的虚机行为分析系统
CN116311214B (zh) 车牌识别方法和装置
US20210327041A1 (en) Image based novelty detection of material samples
CN117596058A (zh) 网络信息的安全保护系统及方法
CN103258209A (zh) 基于三阶统计特征和组合分类器的数字图像篡改盲检测方法
CN113313092A (zh) 手写签名识别方法、理赔自动化处理方法、装置和设备
WO2014198055A1 (en) Image processing including adjoin feature based object detection, and/or bilateral symmetric object segmentation
CN112257646B (zh) 一种商品检测方法、装置、电子设备及存储介质
CN112308061A (zh) 一种车牌字符排序方法、识别方法及装置
Dong et al. A new steganalysis paradigm based on image retrieval of similar image-inherent statistical properties and outlier detection
CN114120122B (zh) 基于遥感影像的灾损识别方法、装置、设备及存储介质
JP5882259B2 (ja) 信号処理装置、方法、及びプログラム
CN117597885A (zh) 散列生成设备、散列确定设备和系统
CN114612686A (zh) 图像识别方法、装置、电子设备和可读存储介质

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2018528057

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20187017248

Country of ref document: KR

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2017882272

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2017882272

Country of ref document: EP

Effective date: 20180722

ENP Entry into the national phase

Ref document number: 2017389535

Country of ref document: AU

Date of ref document: 20170630

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE