CN111860268A - Counterfeit image detection and identification method based on machine learning - Google Patents
Counterfeit image detection and identification method based on machine learning Download PDFInfo
- Publication number
- CN111860268A CN111860268A CN202010667514.8A CN202010667514A CN111860268A CN 111860268 A CN111860268 A CN 111860268A CN 202010667514 A CN202010667514 A CN 202010667514A CN 111860268 A CN111860268 A CN 111860268A
- Authority
- CN
- China
- Prior art keywords
- image
- machine learning
- detected
- detecting
- tampered
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000010801 machine learning Methods 0.000 title claims abstract description 32
- 238000001514 detection method Methods 0.000 title claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 23
- 230000002159 abnormal effect Effects 0.000 claims abstract description 21
- 239000013598 vector Substances 0.000 claims abstract description 18
- 239000002131 composite material Substances 0.000 claims description 6
- 238000000513 principal component analysis Methods 0.000 claims description 6
- 238000007637 random forest analysis Methods 0.000 claims description 6
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 239000011159 matrix material Substances 0.000 description 8
- 238000012549 training Methods 0.000 description 6
- 230000035945 sensitivity Effects 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000005242 forging Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010348 incorporation Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/40—Spoof detection, e.g. liveness detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biomedical Technology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a counterfeit image detection and identification method based on machine learning, which comprises the following steps: firstly, acquiring an image to be detected; then, extracting the image characteristics of the image to be detected to generate the characteristic vector of the image to be detected; and finally, detecting whether the image to be detected is an abnormal image or not through a machine learning algorithm for detecting the abnormal image according to the characteristic vector. Meanwhile, whether the image to be detected is the synthesized image is detected through a machine learning algorithm for detecting the synthesized image, so that the forged image can be detected without a large amount of computer resources and time resources, the detection effect is good, and the type of the forged image can be determined so as to perform targeted processing on the forged image.
Description
Technical Field
The invention relates to the field of methods or devices for identification by using electronic equipment, in particular to a counterfeit image detection and identification method based on machine learning.
Background
In this era of advanced technology, digital images have become an important part of human life, and are used in personal entertainment or medicine. In huge digital images, how to confirm the authenticity of an image has become one of the important issues. From a medical point of view, counterfeiting a medical image can lead to a situation where a doctor misdiagnoses a patient; while politically counterfeiting an image can cause erroneous information to be disseminated, it is therefore seen that detecting a counterfeit digital image is extremely important.
Counterfeit images can be basically divided into three major categories: 1) abnormal images, i.e. images that differ greatly from most other images; 2) a composite image, i.e. an image synthesized by a computer; 3) and (4) tampering the image, namely adding things which do not exist originally into a certain image.
The existing technology for detecting the forged image comprises two methods, namely an active method and a passive method. Active methods refer to the incorporation of a digital watermark into an image, which means that the image has not been modified as long as the watermark is detected as having not been tampered with or damaged. This active method-based detection requires watermarking of the image and ensures that the original image has not been tampered with, and therefore this method can only be used if the authenticity of the image is fully confirmed.
Passive methods analyze certain features in the digital image to determine if the image is counterfeit. Passive methods, while more difficult to practice than active methods, are advantageous in practice because they do not require a priori knowledge of the image. Most of the current methods for passively detecting image forgery are based on deep learning algorithm.
Although the depth learning algorithm can detect whether an image belongs to a forged image, the computer and time resources required for training a depth learning model are huge, and it is difficult to determine the forging mode of the image, i.e. determine whether the forged image belongs to an abnormal image, a synthesized image, a tampered image or a combination of the abnormal image, the synthesized image and the tampered image, so that it is difficult to correspondingly process the forged image according to the forging mode of the image.
Disclosure of Invention
In order to solve the above technical problems, the present invention provides a method for detecting and identifying a counterfeit image based on machine learning, which respectively detects the abnormal, synthetic and falsified situations of an image by 3 different machine learning algorithms to determine the counterfeit image and the type of the counterfeit image.
The technical scheme is as follows:
the method for detecting and identifying the forged image based on the machine learning is provided, and in a first implementation mode, the method comprises the following steps:
acquiring an image to be detected;
extracting image features of the image to be detected and constructing feature vectors;
and respectively adopting three different machine learning algorithms to detect the image to be detected according to the characteristic vectors, wherein the three machine learning algorithms are respectively used for detecting an abnormal image, a synthetic image and a tampered image.
In combination with the first implementable manner, in the second implementable manner, the extracted image features include a mean value, a standard deviation, a skewness, a kurtosis, an energy, an entropy, and a smoothness of the image to be detected.
With reference to the second implementable manner, in a third implementable manner, the image feature further includes: after the image to be detected is divided into a plurality of sub-images, the mean value, the standard deviation, the skewness, the kurtosis, the energy, the entropy and the smoothness of each sub-image.
With reference to the second or third implementable manner, in a fourth implementable manner, the image feature further includes: and after the image to be detected is processed by a Gabor filter, extracting the mean value, the standard deviation, the skewness and the kurtosis.
With reference to any one of the first to fourth implementable manners, in a fifth implementable manner, the abnormal image is detected using a robust principal component analysis algorithm.
With reference to any one of the first to fifth implementable manners, in a sixth implementable manner, the composite image is detected using a support vector machine classifier.
With reference to any one of the first to sixth implementable manners, in a seventh implementable manner, the tampered image is detected by using a random forest classifier.
With reference to any one of the first to seventh implementable manners, in an eighth implementable manner, the method further includes:
separating the tampered image into N blocks of pixels a x a;
calculating a noise level for each pixel block;
clustering all pixel blocks based on the noise level;
and determining a tampered part of the tampered image according to the number of the classified pixel blocks.
With reference to the eighth implementable manner, in a ninth implementable manner, the method further includes:
dividing each n x n pixel block into M sub-pixel blocks;
calculating a noise level for each sub-pixel block;
determining a final noise level for each sub-pixel block based on the noise level for each pixel block and the noise level for each sub-pixel block;
clustering all sub-pixel blocks based on the final noise level;
and determining a tampered part of the tampered image according to the number of the classified pixel blocks.
In combination with the eighth or ninth implementable manner, in the tenth implementable manner, clustering is performed by using a multi-class hierarchical clustering algorithm.
Has the advantages that: the counterfeit image detection and identification method based on machine learning can detect the counterfeit image without a large amount of computer resources and time resources, has good detection effect, and can determine the type of the counterfeit image so as to perform targeted processing on the counterfeit image.
Drawings
FIG. 1 is a flow chart of the detection method of the present invention;
FIG. 2 is a flow chart of a method for detecting an abnormal image;
FIG. 3 is a flow chart of a method for detecting a composite image;
FIG. 4 is a flow chart of a method of detecting tampering with an image;
FIG. 5 is a flow chart of locating a tampered portion;
FIG. 6 is a diagram illustrating the results of anomaly, synthetic and tamper total data set distribution analysis;
FIG. 7 is a diagram of anomaly, synthetic and tampered total data set performance indicators;
fig. 8 is a schematic diagram of a tamper localization result.
Detailed Description
The invention is further illustrated by the following examples and figures.
Fig. 1 is a flow chart of a method for detecting and identifying a counterfeit image based on machine learning, the method comprising:
and 3, detecting the image to be detected by adopting three different machine learning algorithms respectively according to the characteristic vectors, wherein the three machine learning algorithms are used for detecting an abnormal image, a synthesized image and a tampered image respectively.
Specifically, after the image to be detected is obtained, the image to be detected can be converted into a gray image, the existing bilateral filtering technology and contrast limitation are adopted to perform denoising and contrast enhancement processing on the gray image, image features are extracted from the processed gray image, and feature vectors corresponding to the image to be detected are generated.
And finally, detecting whether the image to be detected is an abnormal image or not through a machine learning algorithm for detecting the abnormal image according to the characteristic vector. Meanwhile, whether the image to be detected is the synthesized image or not can be detected through a machine learning algorithm for detecting the synthesized image, and whether the image to be detected is the tampered image or not can also be detected through a machine learning method for detecting the tampered image.
Therefore, the type of the counterfeit image can be determined, and if the three machine learning algorithms do not detect that the image to be detected is not an abnormal, synthesized or counterfeit image, the image to be detected is a real image and is not a counterfeit image.
In this embodiment, preferably, the extracted image features include a mean value, a standard deviation, a skewness, a kurtosis, an energy, an entropy, and a smoothness of the grayscale image, and the mean value, the standard deviation, the skewness, the kurtosis, the energy, the entropy, and the smoothness of each sub-image extracted after the grayscale image of the to-be-detected image is divided into 9 sub-images. And the mean value, the standard deviation, the skewness and the kurtosis are extracted from the processed filtered image after the gray image is processed by a Gabor filter. In this embodiment, the kernel function of the Gabor filter is:
x′＝xcosθ+ysinθ，y′＝-xcosθ+ysinθ；
λ is the wavelength, θ is the angle, ψ is the phase difference, σ is the standard deviation of gaussian envelope, γ is the spatial aspect ratio, and in this example, λ is (2, 2.5, 3, 3.5) and θ is (0 °, 60 °, 120 °, 180 °, 240 °, 300 °), and 24 different filtered images can be obtained by adjusting the parameters of the kernel function. And extracting the mean value, the standard deviation, the skewness and the kurtosis of each filtered image.
In conjunction with the features extracted from the grayscale image, all sub-images, and all filtered images, a feature vector can be generated that includes 166 features.
In this embodiment, preferably, as shown in fig. 2, detecting the abnormal image by using a robust principal component analysis algorithm specifically includes:
step 1-1, generating an n multiplied by p data matrix according to the characteristic vector;
step 1-2, projecting the data matrix to a dimension of at most n-1;
step 1-3, calculating a covariance matrix of a data matrix and k main components which need to be reserved for better fitting data on a dimension space of n-1;
step 1-4, projecting each image data point in the data matrix to the k dimensional space and calculating k corresponding nonzero eigenvalues and eigenvectors to form a robust principal component, wherein the image data point is each image data of the data matrix in the step 1-1, namely each row of the data matrix;
Step 1-5, calculating a robust score distance (Scoredistance) and an Orthogonal distance (Orthogonal distance) corresponding to each data point by using a robust principal component;
step 1-6, calculating a threshold value based on the robust score distance and the orthogonal distance of all image data points, and specifically comprising the following steps:
step 1-6-1, use
Calculating a threshold for the robust score distance, wherein
Scoring a 97.5% quantile of distance value chi-square distribution for all robustness;
step 1-6-2, use
Calculating a threshold for the orthogonal distance, where μ, σ and z0.975The mean, standard deviation and 97.5% quantile of the positive Taiwan distribution, respectively.
Step 1-7, based on the robust score distance and the threshold of the orthogonal distance, categorizes each image data point into one of the following 4 possibilities: 1) a normal data point when both the robust score distance value and the orthogonal distance value are less than the threshold, 2) a lever data point when the robust score distance value is greater than the threshold and the orthogonal distance value is good when less than the threshold, 3) an orthogonal outlier when the robust score distance value is less than the threshold and the orthogonal distance value is greater than the threshold, and 4) a bad lever data point when both the robust score distance value and the orthogonal distance value are greater than the threshold;
and 1-8, defining bad lever data points as abnormal images.
In this embodiment, preferably, as shown in fig. 3, detecting the composite image by using a support vector machine classifier specifically includes:
step 2-1, collecting feature data and mark data of each sample image in a synthetic image sample set, and constructing a training data set, wherein the mark data comprises whether each sample image is a synthetic image, if the sample is the synthetic image sample, the mark data is 1, if the sample is a normal image sample, the mark data is 0, the synthetic image sample set comprises a plurality of normal image samples and synthetic image samples, and the feature data is the same as the feature vector and comprises 166 features;
step 2-2, searching a hyperplane w capable of separating data points of the normal image and the synthetic image according to the characteristic space in which the characteristic data is positionedTXi+ b is 0, and the distance from any sample point to the hyperplane is greater than or equal to 1 to satisfy the following condition:
wTxi+b≥1
where w and b are the normal vector and intercept of the hyperplane, XiIs the ith data point;
2-3, selecting an optimal hyperplane from all hyperplanes meeting the conditions, wherein the interval boundary distance between data points of the two images can be maximized;
and 2-4, identifying whether the image to be detected is a synthetic image or not through the feature vector corresponding to the image to be detected based on the trained support vector classifier.
In this embodiment, as shown in fig. 4, the detecting the tampered image by using a random forest classifier specifically includes:
step 3-1, collecting feature data and marking data of each image sample in a tampered image sample set, and constructing a training data set, wherein the marking data comprises whether each sample image is a tampered image, the marking data is 1 if the image sample is tampered, the marking data is 0 if the image sample is a normal image sample, the tampered image sample set comprises a plurality of normal image samples and tampered image samples, the feature data is the same as the feature vector and comprises 166 features;
step 3-2, constructing a plurality of subdata sets according to the training data set, wherein sample data in the subdata sets are obtained by sampling returned from the training data set, and the characteristic data quantity in the subdata sets is the same as that of the training data set;
3-3, respectively constructing a corresponding decision tree through each sub data set, and in the process of constructing the decision tree, randomly selecting less than 166 characteristics from the sub data sets when each node is split, and selecting the optimal characteristics from the sub data sets according to the Kearny coefficient to split;
step 3-4, combining the random forest classifier according to all the constructed decision trees;
And 3-5, identifying whether the image to be detected is a tampered image or not through the feature vector corresponding to the image to be detected based on the trained random forest classifier.
After determining that the image to be detected is a tampered image, in this embodiment, as shown in fig. 5, the following method may also be used to locate the tampered portion in the tampered image:
step 4-1, dividing the tampered image into N pixel blocks a x a;
4-2, calculating the noise level of each pixel block;
4-3, clustering all pixel blocks based on the noise level;
and 4-4, determining a tampered part of the tampered image according to the number of each classified pixel block.
The method specifically comprises the following steps:
first, the tampered image may be divided into N64 x 64 pixel blocks.
The noise level of each pixel block can then be calculated by a principal component analysis algorithm. The method specifically comprises the following steps:
step 4-4-1, dividing each pixel block into a plurality of overlapped small pixel blocks of 5 × 5;
step 4-4-2, calculating the eigenvalue alpha of the covariance matrix of each small pixel blocki；
Step 4-4-3, gradually discarding the small pixel blocks with the largest characteristic value until the following conditions are met:
wherein alpha is7And alphaminRespectively the 7 th smallest eigenvalue and the minimum of all eigenvalues, the noise level of the pixel block, initially defined as C 0Q(p0) Wherein Q (p)0) Is the p-th of the characteristic value0Quantile, in this example, C0＝3.1，p0＝0.0005；
Step 4-4-4, calculating the noise level of the entire pixel block
And returning to the step 4-4-3 until the difference value between the noise levels of the pixel blocks obtained by two successive calculations is less than 1 multiplied by 10-6。
And finally, clustering each pixel block by a multi-class hierarchical clustering algorithm and the difference statistics based on the noise level to obtain a plurality of classification sets, wherein the pixel block in the classification set with the least pixel blocks is the tampered part.
Specifically, first, each pixel block is regarded as an independent node and the similarity of each pair of nodes is calculated, and in the present embodiment, the similarity index used is the euclidean distance. The corresponding node pairs are then sorted and aggregated according to similarity from high to low. This is repeated until the desired number of clusters is reached.
In this embodiment, to determine the optimal cluster number, it may be assumed that the cluster number is between 1 and 5, then calculate the difference statistic and the Silhouette statistic of each cluster number, and select the cluster number having the largest difference statistic and the Silhouette statistic as the optimal cluster number.
In order to more accurately locate the tampered part, in this embodiment, after the tampered part is determined by the above method, the tampered part in the tampered image is further located by the following method:
Step 5-1, dividing each n × n pixel block into M M × M sub-pixel blocks;
step 5-2, calculating the noise level of each sub-pixel block;
step 5-3, determining the final noise level of each sub-pixel block according to the noise level of each pixel block and the noise level of each sub-pixel block;
step 5-4, clustering all sub-pixel blocks based on the final noise level;
and 5-5, determining a tampered part of the tampered image according to the number of each classified sub-pixel block.
Specifically, first, each 64 × 64 pixel block may be subdivided into M32 × 32 sub-pixel blocks.
Then, the noise level of each sub-pixel block is calculated, again using a principal component analysis algorithm.
Then, a final noise level for each sub-pixel block is calculated based on the noise level for each pixel block and the noise level for each sub-pixel block. Wherein the noise level of the sub-pixel block previously belonging to the normal portion is the same as the noise level of the pixel block, and the noise level σ of the sub-pixel block previously belonging to the pixel block corresponding to the tampered portion is 0.8 σ64+0.2σ32。
And finally, based on the final noise level of each sub-pixel block, clustering each sub-pixel block by adopting a multi-class hierarchical clustering algorithm according to Silhouette statistics to obtain a classification set of a plurality of sub-pixel blocks, wherein the sub-pixel block in the classification set with the least number of sub-pixel blocks is the tampered part.
For the detection of abnormal images, the steady principal component analysis can have good results at different percentages of abnormality degrees. The present invention can obtain an accuracy (precision) of 0.500, a recall (recall) of 1.000, an F1 value (F1 score) of 0.667, and a Matthews correlation coefficient (Matthews correlation coefficient) of 0.704 in an abnormality degree of 1%. In the case of 5%, 10%, and 25%, the four indices are 1.000, indicating that all the abnormal pictures can be detected even in the case of a large number of abnormal pictures.
The detection performance analysis for the composite images was performed by calculating the sensitivity (sensitivity), specificity (specificity) and accuracy (accuracuracy) of crossing the validation image set (9959 images) and the test image set (2489 images) at 10 × 10 folds. The final results show that the sensitivity, specificity and accuracy of the data set are respectively 99.97% (95% confidence interval: 99.92-100%), 99.99% (95% confidence interval: 99.97-100%) and 99.98% (95% confidence interval: 99.94-100%) by using a cubic polynomial support vector machine, and the results of the data set are 100% (95% confidence interval: 100 and 100%).
There were 37549 and 9387 images, respectively, for the 10 x 10 fold cross validation image set and the test image set of the tampered images. The sensitivity, specificity and accuracy of the final random forest classifier in the validation set were 96.97% (95% confidence interval: 96.89-97.04%), 98.11% (95% confidence interval: 98.01-98.21%) and 97.54% (95% confidence interval: 97.48-97.60%), respectively, while the results in the test set were 97.38% (95% confidence interval: 95.82-97.79%), 97.78% (95% confidence interval: 97.33-98.20%) and 97.58% (95% confidence interval: 97.22-97.88%).
As shown in fig. 6, three types of image falsifications (anomaly, composition, and falsification) are finally put into an image set, of which there are 6 image types: exceptions (anomalouses, A) and non-exceptions (Normal, N), Synthetic (Synthetic, S) and non-Synthetic (Real, R), falsified (Tampered, T) and non-falsified (Pristine, P) form 8 combinations. The three-layer machine learning method used by the invention can classify under the condition that three kinds of image forgery exist and define a bad lever data point (the upper right area of figure 6) as an abnormal image. As shown in fig. 7, the present invention also obtains the precision, recall, F1 values, and Matthews correlation coefficients of 0.752, 0.840, 0.776, and 0.756 in the image sets of the above 1000 image samples.
As shown in fig. 8, tampering, whether small (fig. 8 first row) or large (fig. 8 second row), can be detected by the present invention, and 32 x 32 pixel squares are used to show the tampered site.
In summary, the performance index when detecting abnormalities, synthesizing and tampering images alone is 95% or more. Besides, various different counterfeits can be detected simultaneously, the tampered place of the image can be positioned, and the function can enable a user to perform downstream analysis.
Finally, it should be noted that the above-mentioned description is only a preferred embodiment of the present invention, and those skilled in the art can make various similar representations without departing from the spirit and scope of the present invention.
Claims (10)
1. A counterfeit image detection and identification method based on machine learning is characterized by comprising the following steps:
acquiring an image to be detected;
extracting image features of the image to be detected and constructing feature vectors;
and respectively adopting three different machine learning algorithms to detect the image to be detected according to the characteristic vectors, wherein the three machine learning algorithms are respectively used for detecting an abnormal image, a synthetic image and a tampered image.
2. A method for detecting and identifying counterfeit images based on machine learning as claimed in claim 1, wherein the extracted image features comprise a mean, a standard deviation, a skewness, a kurtosis, an energy, an entropy and a smoothness of the image to be detected.
3. The method for detecting and identifying a counterfeit image based on machine learning according to claim 2, wherein the image features further include: after the image to be detected is divided into a plurality of sub-images, the mean value, the standard deviation, the skewness, the kurtosis, the energy, the entropy and the smoothness of each sub-image.
4. A method for detecting and identifying a counterfeit image based on machine learning according to claim 2 or 3, wherein the image features further comprise: and after the image to be detected is processed by a Gabor filter, extracting the mean value, the standard deviation, the skewness and the kurtosis.
5. A method for detecting and identifying forged images based on machine learning according to any of claims 1-4, wherein a robust principal component analysis algorithm is used to detect the abnormal images.
6. A method for detecting and identifying counterfeit images based on machine learning according to any of claims 1-5, wherein a support vector machine classifier is used to detect the composite image.
7. A method for detecting and identifying forged images based on machine learning as claimed in any of claims 1-6, wherein a random forest classifier is used to detect the tampered images.
8. A method for detecting and identifying a counterfeit image based on machine learning according to any one of claims 1 to 7, further comprising:
separating the tampered image into N blocks of pixels a x a;
calculating a noise level for each pixel block;
clustering all pixel blocks based on the noise level;
and determining a tampered part of the tampered image according to the number of the classified pixel blocks.
9. A method for detecting and identifying a counterfeit image based on machine learning according to claim 8, further comprising:
dividing each n x n pixel block into M sub-pixel blocks;
calculating a noise level for each sub-pixel block;
determining a final noise level for each sub-pixel block based on the noise level for each pixel block and the noise level for each sub-pixel block;
clustering all sub-pixel blocks based on the final noise level;
and determining a tampered part of the tampered image according to the number of each classified sub-pixel block.
10. A method for detecting and identifying a counterfeit image based on machine learning according to claim 8 or 9, wherein: and clustering by adopting a multi-class hierarchical clustering algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010667514.8A CN111860268A (en) | 2020-07-13 | 2020-07-13 | Counterfeit image detection and identification method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010667514.8A CN111860268A (en) | 2020-07-13 | 2020-07-13 | Counterfeit image detection and identification method based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111860268A true CN111860268A (en) | 2020-10-30 |
Family
ID=72984379
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010667514.8A Pending CN111860268A (en) | 2020-07-13 | 2020-07-13 | Counterfeit image detection and identification method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111860268A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329596A (en) * | 2020-11-02 | 2021-02-05 | 中国平安财产保险股份有限公司 | Target damage assessment method and device, electronic equipment and computer-readable storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104408728A (en) * | 2014-12-03 | 2015-03-11 | 天津工业大学 | Method for detecting forged images based on noise estimation |
| US20160071264A1 (en) * | 2014-09-06 | 2016-03-10 | RaPID Medical Technologies, LLC | Medical image dectection system and method |
| CN109829889A (en) * | 2018-12-27 | 2019-05-31 | 清影医疗科技（深圳）有限公司 | A kind of ultrasound image processing method and its system, equipment, storage medium |
| CN109961093A (en) * | 2019-03-07 | 2019-07-02 | 北京工业大学 | A kind of image classification method based on many intelligence integrated studies |
-
2020
- 2020-07-13 CN CN202010667514.8A patent/CN111860268A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160071264A1 (en) * | 2014-09-06 | 2016-03-10 | RaPID Medical Technologies, LLC | Medical image dectection system and method |
| CN104408728A (en) * | 2014-12-03 | 2015-03-11 | 天津工业大学 | Method for detecting forged images based on noise estimation |
| CN109829889A (en) * | 2018-12-27 | 2019-05-31 | 清影医疗科技（深圳）有限公司 | A kind of ultrasound image processing method and its system, equipment, storage medium |
| CN109961093A (en) * | 2019-03-07 | 2019-07-02 | 北京工业大学 | A kind of image classification method based on many intelligence integrated studies |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329596A (en) * | 2020-11-02 | 2021-02-05 | 中国平安财产保险股份有限公司 | Target damage assessment method and device, electronic equipment and computer-readable storage medium |
| CN112329596B (en) * | 2020-11-02 | 2021-08-24 | 中国平安财产保险股份有限公司 | Target damage assessment method and device, electronic equipment and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gao et al. | Automatic change detection in synthetic aperture radar images based on PCANet | |
| Ryu et al. | Rotation invariant localization of duplicated image regions based on Zernike moments | |
| Prakash et al. | Detection of copy-move forgery using AKAZE and SIFT keypoint extraction | |
| CN107622489B (en) | Image tampering detection method and device | |
| Al-Qershi et al. | Evaluation of copy-move forgery detection: datasets and evaluation metrics | |
| CN112818862A (en) | Face tampering detection method and system based on multi-source clues and mixed attention | |
| Raghavendra et al. | Presentation attack detection algorithms for finger vein biometrics: A comprehensive study | |
| Aghdaie et al. | Detection of morphed face images using discriminative wavelet sub-bands | |
| Wu et al. | Identifying computer generated graphics via histogram features | |
| CN103942540A (en) | False fingerprint detection algorithm based on curvelet texture analysis and SVM-KNN classification | |
| CN111275070B (en) | Signature verification method and device based on local feature matching | |
| CN111768368B (en) | Image area copying and tampering detection method based on maximum stable extremal area | |
| CN112488211A (en) | Fabric image flaw classification method | |
| Tahaoglu et al. | Improved copy move forgery detection method via L* a* b* color space and enhanced localization technique | |
| Yuan et al. | Fingerprint liveness detection using histogram of oriented gradient based texture feature | |
| Skytte et al. | Evaluation of yogurt microstructure using confocal laser scanning microscopy and image analysis | |
| Kiruthika et al. | Image quality assessment based fake face detection | |
| Warif et al. | A comprehensive evaluation procedure for copy-move forgery detection methods: results from a systematic review | |
| CN117558011B (en) | Image text tampering detection method based on self-consistency matrix and multi-scale loss | |
| Kaur et al. | A deep learning framework for copy-move forgery detection in digital images | |
| Sabeena et al. | Convolutional block attention based network for copy-move image forgery detection | |
| CN111860268A (en) | Counterfeit image detection and identification method based on machine learning | |
| CN117636421A (en) | Face deep pseudo detection method based on edge feature acquisition | |
| CN117373136A (en) | Face counterfeiting detection method based on frequency mask and attention consistency | |
| Makandar et al. | Transform Domain Techniques combined with LBP for Tampered Image Detection using Machine Learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201030 |