CN113920434A - Image reproduction detection method, device and medium based on target - Google Patents
Image reproduction detection method, device and medium based on target Download PDFInfo
- Publication number
- CN113920434A CN113920434A CN202111253036.7A CN202111253036A CN113920434A CN 113920434 A CN113920434 A CN 113920434A CN 202111253036 A CN202111253036 A CN 202111253036A CN 113920434 A CN113920434 A CN 113920434A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- reproduction
- model
- copied
- 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
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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a device and a medium for detecting image reproduction based on a target, wherein the method comprises the following steps: acquiring an image to be detected, detecting the image by adopting a preset first image reproduction detection model, and determining whether the image to be detected is an image without a target, a reproduced image of the target or a non-reproduced image of the target according to a result; and cutting the image with the result of non-reproduction by using a preset image preprocessing method, extracting the obtained new image by using a preset image characteristic extraction method, inputting the extracted characteristic into a preset second image reproduction detection secondary classification model for further prediction, and determining whether the image is a reproduction image according to the result. The method solves the problem that the existing method can not directly and automatically identify whether the target in the image is the reproduction or not through two-stage reproduction detection. The first stage detects the target reproduction image with poor quality and obvious screen frame or mole pattern, and the second stage detects the target reproduction image with high quality and no obvious screen frame or mole pattern.
Description
Technical Field
The invention relates to the field of image processing, in particular to a method, a device and a medium for detecting image reproduction based on a target.
Background
With higher and higher informatization degree, more and more application scenes need to identify the authenticity of images of commodities, certificates and the like by photographing, particularly in special industries such as fast-moving consumer goods retail industry, image anti-counterfeiting industry and the like. However, in the current image recognition method, it is assumed that the image itself is completely reproduced, and neglecting that the commodity object or the certificate object in the image to be detected may only occupy a part or even a small part of the whole image. This means that only a small part of the image may be rendered on-screen for the entire image and is mostly natural, thus having a major impact on those algorithms that attempt to detect whether the image is rendered by directly using features of the entire rendered image, such as moire patterns. In addition, there may be some images, which contain the non-copied target and other copied non-targets, which will also cause erroneous judgment caused by using most current copy recognition algorithms to detect whether the goods or the certificate itself is copied. Meanwhile, in the field of image anti-counterfeiting, some counterfeiters or attackers may use the image of the type to attack an anti-counterfeiting system to achieve the purpose of being unaffordable.
Therefore, how to detect whether the commodity target or the certificate target is copied or not in a full-automatic and accurate manner aiming at the common image is very important for related enterprises such as fast-moving consumer goods retail industry, image anti-counterfeiting enterprises and the like.
However, in the existing copy detection method, although there is a method of extracting a target by using a target detection method and the like, the copy detection is performed on the target area by using the corresponding features. However, the accuracy of final reproduction detection is affected by the extracted targets possibly being small, and the reproduction detection method is not practical due to the fact that image target labeling and model training in the target extraction model are time-consuming.
Disclosure of Invention
In order to effectively solve the above problems, embodiments of the present invention provide a method, an apparatus, and a medium for detecting image reproduction of a target, which can detect whether a target in a general image is reproduced by combining two stages of image reproduction detection methods. The method can fully automatically, accurately and efficiently identify whether the image to be detected contains the target and whether the target is copied, whether the image is a low-quality copied image containing a screen frame or moire fringes or whether the copied image is a high-quality copied image without the screen frame or moire fringes.
Only a single target is provided, and for the detection of the reproduction of a plurality of targets, two types can be added to each target in the training set in the first image reproduction detection model, wherein one type is a reproduction image containing a new target, and the other type is a non-reproduction image sample set containing the new target, namely if n targets exist, the first training image sample set has 2n +1 types.
The technical scheme adopted by the embodiment of the specification is as follows: the embodiment of the specification provides an image copying detection method based on a target, which comprises the following steps:
the method comprises the steps of obtaining an image to be detected, detecting the image according to a preset first image reproduction detection model, carrying out preliminary image reproduction detection, and detecting a target reproduction image with an obvious screen frame or clear mole lines at the stage.
The method comprises the steps that the first image copying detection model identifies the copied image with a screen frame or clear moire accurately, but the copied image with few frames or no frames and no clear moire is easily identified as a non-copied image, so that the edge part of the image which is predicted to be non-copied by the first image copying detection model is cut off, and the rest images are further copied and detected by the second image copying detection model.
Identifying the first image reproduction detection model as an image to be detected with a non-reproduction target, and performing edge cutting by adopting a preset image preprocessing method to obtain a new image; the reason for clipping is that a clear copied image with only a small part of frame at the edge of a part of image may be recognized as a non-copied image in the first image copying detection model, and clipping the small part of frame by the image preprocessing clipping method is beneficial to improving the accuracy of prediction of the second image copying detection model.
And inputting the image obtained by preprocessing into a preset second image reproduction detection model for further reproduction identification.
Further, the method for recognizing image reproduction based on a target is characterized in that the acquiring a plurality of first image sets to be trained specifically includes:
acquiring a plurality of original images, wherein the original images comprise three types, namely, a target in the image is copied, the image does not contain the target, and the target in the image is not copied;
the target in the image is that the copied image can be a frame containing screen equipment such as a mobile phone or a computer or a frame containing obvious moire;
the image without the target in the image can be any image, but does not contain the target;
the objects in the image that are not in the copied training image must not be copied, but other copied components are allowed in the image.
Furthermore, in the to-be-trained image set, in order to improve the classification accuracy, it is necessary to consider adding some confrontation images in each type of image, for example, adding some images including a mobile phone frame or a computer frame in the image without the target type. Some image target parts are added in a non-reproduction image class containing targets in the image and are not reproduced, but other parts in the image may contain other reproduction parts.
Further, in the first image duplication detection method, the first image set to be trained is input into a three-classification convolutional neural network for training, so as to obtain the first image duplication detection model.
Further, the image duplication detection method based on the target is characterized in that,
the image preprocessing is performed on the non-copied image identified as containing the target in the first image copying detection model, and the main purpose of the preprocessing is that the probable edge part in the image detected as not being copied in the first image copying detection model contains less information such as frames. The edge portion is trimmed away, but at the same time the target is retained without loss and the trim cannot be too large.
Further, the target-based image reproduction detection method is characterized in that the preprocessed image is input into a second image reproduction detection model for further identification.
Further, in the second image reproduction detection classification method, characterized in that,
and acquiring a plurality of second image sample sets to be trained, wherein the original images comprise two types, one type is a natural image containing a target, the other type is a copied image containing the target, and the images copied by the second type of target are clear and contain no screen frame or copied images with few or no moire fringes.
Further, in the second image reproduction detection method, characterized in that,
extracting image characteristics of two types of images to be trained to obtain an image characteristic data set;
and training the extracted image feature set by adopting a Support Vector Machine (SVM) method to obtain the second image copying detection binary classification model.
On the basis of the method embodiment, the invention correspondingly provides an apparatus embodiment.
Another embodiment of the present invention provides a device for detecting image duplication based on a target, including: the device comprises a first image reproduction detection module, an image preprocessing module and a second image reproduction detection module.
The first image reproduction detection module is used for detecting whether the image to be detected contains a target or not and whether the contained target is a reproduction image or not, and the detection target reproduction image possibly has obvious frame or moire information. For a copied image containing a target and no obvious frame or no obvious moire, the first image copying detection model tends to identify the copied image as a non-copied image containing the target.
Furthermore, the first image copying detection module comprises a first image sample set acquisition subunit, a sample set division subunit and a model training subunit.
The first image sample set obtaining subunit is configured to obtain a plurality of images to be trained, where the images to be trained include a target or no target.
The sample set dividing subunit is configured to divide the image sample set into 8: 1: 1, 80% for model training, 10% for model validation, and 10% for model testing.
The image preprocessing module is used for removing the frame of the image which is identified as the non-reproduction clear image containing the target in the first stage and has little or no obvious frame and possibly has the edge, so that the precision of the second image reproduction detection model is improved.
And the second image copying detection module is used for further copying and detecting the preprocessed image, and is used for detecting the image without the moire pattern and the image without the frame containing the target.
Further, the second image copying detection module comprises a second image sample set acquisition subunit, an image feature extraction subunit and a classification model training subunit.
The second image sample set acquisition subunit is used for acquiring clear molar-pattern-free and frame-free reproduction images with targets and natural images with targets.
And the image feature extraction subunit extracts image features. Further, graying the image to obtain a grayscale image, performing image feature extraction of relevant statistics on the obtained grayscale image, and taking the obtained image feature set as an actual training set of the second image reproduction detection model.
And classifying the extracted feature data set by the second image reproduction detection method by adopting an SVM (support vector machine) method to obtain a two-classification model.
On the basis of the above-described method embodiments, the present invention provides corresponding medium item embodiments.
An embodiment of the present invention provides a storage medium, which includes a stored computer program, where the computer program is executed to implement the target-based image duplication detection method according to any one of claims 1 to 13.
Compared with the prior art, the technical scheme of the embodiment of the invention has the following beneficial effects:
according to the image copying detection method, device and medium based on the target, whether the image to be detected contains the target or not and whether the target is copied or not are determined firstly according to the preset first image copying detection method for the image to be detected. The detection method can accurately identify the copied image containing the screen frame or the moire fringes. In order to further improve the reproduction detection accuracy, the image which is detected to be the non-reproduction image is input into a second image reproduction detection method for further reproduction detection by adopting preset image preprocessing, the clear frameless reproduction image and the natural image are classified and identified, and finally whether the image to be detected is the image which is detected to be the reproduction image or not is determined.
The embodiment of the invention realizes the full-automatic detection of whether the image to be detected contains the target or not and whether the target image is a copied image or not. The invention divides the image to be detected by the two-stage copying detection method, and can help certificate counterfeiting, fast-disappearing product enterprises and image anti-counterfeiting enterprises to effectively solve the problems in business.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly described below.
Fig. 1 is a schematic flowchart of a target-based image duplication detection method according to an embodiment of the present invention.
Fig. 2 is a schematic structural diagram of an image duplication detection apparatus based on a target according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of three types of samples in a first image capture test model training set according to an embodiment of the present invention, wherein a circular cover is the target.
FIG. 4 is a diagram illustrating pre-processing cropped edges of a sharper rendered image with a screen border at the edge of the image according to an embodiment of the present invention.
FIG. 5 is a schematic diagram of two types of samples in the training set of the second image capture detection model according to an embodiment of the invention.
FIG. 6 is a block diagram of an embodiment of the invention.
Fig. 7 is a flowchart of the first image duplication detection module according to an embodiment of the present invention.
Fig. 8 is a flowchart of the second image duplication detection module according to an embodiment of the present invention.
Detailed Description
The analysis of the prior art shows that in the current image reproduction detection technical scheme, it is generally assumed that a part of a reproduction image to be detected which is reproduced by a screen of a mobile phone or a computer and the like contains a commodity or a certificate target, and no part of a non-reproduction image to be detected is reproduced by the screen, and then reproduction identification is carried out by utilizing the classification algorithm modeling in machine learning by means of image characteristics such as a screen equipment frame, reflection, moire patterns and the like. However, in practical applications, what we need to really detect is whether the goods or certificate object in the image is screen-copied, and do not care about whether the non-goods object part in the image to be detected is copied.
The invention provides a method, a device and a medium for detecting image reproduction based on a target, which can quickly and accurately identify whether the general image contains the target and whether the target part in the image is reproduced.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all embodiments of the present invention. Optionally, the target in this embodiment may be a commodity target or a certificate target, and this embodiment only considers one target, and it is simple to popularize to most targets.
Fig. 1 is a schematic flowchart of a target-based image duplication detection method according to an embodiment; as shown in fig. 2, a method flow for detecting image duplication based on a target according to an embodiment of the present invention includes the following steps.
Step 101: and acquiring an image to be detected and performing copying identification according to a preset first image copying detection model.
Step 102: and identifying the first image reproduction detection method as a reproduction image containing the target, and preprocessing the reproduction image by using a preset image preprocessing method to obtain a new image.
Step 103: and performing feature extraction on the new image obtained by the preset image preprocessing method by using a preset image feature extraction method, and then inputting the extracted image features into the second image reproduction detection model for further reproduction and identification.
For step 101, optionally, the image to be measured may be an image with or without a target in any format, such as JPEG, PNG, and the like, and may be one or more images.
Further, for step 101, the specific construction of the preset first image duplication detection model includes.
A1, acquiring a plurality of first to-be-trained image sample sets: wherein the first image set of samples to be trained comprises three types of images; specifically, an image not including the target, a copied image including the target, and a non-copied image including the target are acquired.
By way of example, an exemplary illustration of three types of images targeted for a circular cover is given in FIG. 3, where FIG. 3(a) is an image without a target; FIG. 3(b) is a reproduction image containing a target; fig. 3(c) is a non-reproduced image containing an object.
Preferably, in the copied image containing the target, a plurality of shooting devices are required to be used for acquiring target images under a plurality of screens, the images contain screen frames in different ranges or Moire patterns in different degrees, and the shooting and screen devices comprise a plurality of mobile phones, flat panels or cameras with different brands and different models; in addition, the background in the captured image should be as rich as possible.
Preferably, some countermeasure images should be added to each category, for example, in the image without the object, a part of the copied image without the object can be added; in the non-reproduced image containing the object, an image containing something else reproduced may be added. In these two types of images, frame information such as a mobile phone or a computer can appear.
A2, as per 8: 1: and 1, randomly dividing the first image sample set to be trained into a training set, a verification set and a test set.
A3, inputting the first image sample set to be trained into a three-classification convolutional neural network for training, and obtaining the first image copying detection classification model.
Preferably, the convolutional neural network model uses a residual neural network ResNet-50.
Preferably, in order to improve the generalization ability of the classification in the training of the model in a3, an image enhancement method is further adopted in the training of the model, wherein the image enhancement method optionally includes image rotation, horizontal and vertical flipping, and the like.
Optionally, in step 102, the image to be detected, which is detected by using the preset first image copying detection model as the image whose target is the non-copied image, is subjected to image processing by using preset image preprocessing, that is, a cutting measure is taken to remove a possible few frame edges. The image to be measured can be cut in small amplitude from top to bottom and from left to right according to the size of the image to be measured in proportion or in a fixed pixel size. Optionally, 100 pixels are cut off from the top, the bottom, the left and the right, or 5% of the width or the height of the image is cut off from the top, the bottom, the left and the right.
For example, as shown in fig. 4, the circular cover object is copied, but the image has no obvious moire and has only a few borders at the edge of the image, and this image is easily recognized as a non-copied image containing the object in the first image copying detection model, so we can use a preset clipping method such as the edge clipping in fig. 4 to remove the peripheral part.
For step 103, optionally, the image feature may be a feature of a grayscale image obtained by graying the image to be detected, which may be a local binary pattern LBP feature or other features that embody an image texture. Wherein, for the RGB image, the following formula can be adopted to calculate the gray value,
wherein R, G, B represent 3 channels of the RGB image, respectively.
For step 103, the specific construction of the second image copying detection binary model includes.
B1, acquiring a plurality of second image sample sets to be trained: the second image sample set to be trained is two types of images, specifically, a target is a copied image and a target is a non-copied image.
For example, see fig. 5, where a circular cover is the target, where fig. 5(a) is a non-replicated image containing the target; fig. 5(b) is a reproduced image containing an object, and it can be seen that the reproduced image has no obvious moire and no screen frame.
In order to improve the accuracy of the second image reproduction detection model, it is preferable that a clear high-quality reproduction image without moire patterns is acquired in the acquisition of an image sample set with a reproduction target, wherein reproduction equipment and a screen are similar to the method for collecting the first image sample set to be trained.
And B2, extracting image features of a second image to be trained, preferably extracting a Local Binary Pattern (LBP) with consistent texture features of a grayed second sample image to be trained, and storing the LBP to a txt file according to categories to obtain a second image feature data set to be trained.
And B3, training the second image feature data set to be trained obtained from the B2 by adopting a classification algorithm to obtain the second image copying detection model. Preferably, the classification algorithm adopts a Support Vector Machine (SVM) method, and the adopted kernel function is a Radial Basis Function (RBF).
For the LBP operator in B2, the original LBP operator only relates to a neighborhood range of 3x3, and in order to adapt to features of different scales, we can choose a multi-scale LBP feature, and based on the improvement of LBP proposed by Ojala et al, we replace the square neighborhood with a circular neighborhood with radius R.
For extracting the LBP features of the consistent pattern in B2, preferably, we use multi-scale LBP features. Optionally, 4 LBP operators are used: LBP8,1、LBP16,2、LBP24,3And LBP24,4To obtain the LBP characteristics. The LBP feature histograms of the consistent patterns obtained by using these four LBP operators are stitched together as the final image feature of step 103.
Wherein LBPP,RThe operator represents the LBP operator containing P sample points within a circular region of radius R, defined as follows:
in the formula (x)c,yc) As a central pixel coordinate, gpIs the gray value of the neighborhood pixel, and gcIs the gray value of the central pixel,
the pixel coordinates of the P points are calculated by the following formula
By implementing the method, the image copying detection based on the target can be realized.
On the basis of the above-described method item embodiment, the invention provides a corresponding apparatus item embodiment.
As shown in fig. 6, an apparatus for detecting image duplication based on a target according to an embodiment of the present invention includes a first image duplication detection module, an image preprocessing module, and a second image duplication detection module.
The first image copying detection module is used for detecting whether the image contains a target or not and detecting whether the target is a copied image or not, and the detection module is mainly characterized in that the copied image of the target containing a screen frame and clear moire fringes can be accurately identified, but the image without the screen frame and the moire fringes can be easily identified as the image with the target being copied. Therefore, in order to further accurately identify such an image, we will perform further image preprocessing and second image duplication detection.
The image preprocessing module is used for cutting the clear image without the moire pattern and with few frames only at the edge of the image, so that the negative influence of the image on the feature extraction in the subsequent second image reproduction detection model is avoided.
And the second image reproduction detection module is used for further detecting a reproduction image which contains a target and is clear, has no screen frame and no moire fringes.
As shown in fig. 7, the first image duplication detection model building module includes a first image sample set obtaining subunit, a sample set dividing subunit, and a model training subunit.
The first image sample set acquisition subunit is used for acquiring a plurality of images to be trained; the images to be trained are images without targets, screen-shot images with targets and non-shot images with targets. The shooting device and the screen device for copying the image should contain different brands and different models, and the copied image in the sample set should contain multiple types, such as all screen frames, partial frames, no frames, and clear and generally clear images with clear moire patterns. In addition, in each type of image set, a part of the confrontation image should be contained, and for example, a reproduction component should be contained in the image set without the target.
The sample set dividing subunit is configured to perform 8: 1: 1 random partitioning, wherein 80% is used for the convolutional neural network model training, 10% is used for the classification model verification, and 10% is used for model testing.
And the model training subunit is used for training the first image sample set by adopting a convolutional neural network model, wherein the adopted convolutional neural network model can be a residual neural network classification model ResNet series, and can also be neural network models such as VGG (variable gradient g) and GoogleLeNet.
Further, the image preprocessing module is configured to perform edge clipping on an image that is predicted to be non-copied as a target in the first image copying detection module, and clip a frame that may exist at an image edge in the image to be detected, so as to reduce an influence on a subsequent prediction result of the second image copying detection module.
Further, as shown in fig. 8, the second image duplication detection model building module includes a second image sample set obtaining subunit, a feature extraction subunit, and a model training subunit.
The second image sample set obtaining subunit is used for obtaining a plurality of images to be trained; wherein the image to be trained comprises a natural image containing the target and a corresponding reproduction image. The copied image is a natural image and is obtained by copying through different shooting equipment and different screen equipment. The copied image is a clear, molar-grain-free and frame-free copied image containing the target.
The feature extraction subunit is configured to extract a local binary pattern LBP feature construction feature data set of the images in the second image sample set.
And the model training subunit is used for inputting the acquired image characteristic data set into an SVM classification algorithm to obtain a second image reproduction detection model.
It should be noted that the above-mentioned embodiments of the apparatus correspond to the above-mentioned embodiments of the method.
On the basis of the above embodiment of the method item, there is correspondingly provided a medium containing a stored computer program, wherein the computer program runs a method for implementing the target-based image duplication detection according to any one of the method items of the present invention.
The storage medium is a computer-readable storage medium, wherein the module/unit integrated with the object-based image duplication detection apparatus may be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as a stand-alone product.
The computer-readable medium may include: any entity or device capable of carrying the computer program code, a removable disk, a U disk, computer Memory, Read-Only Memory (ROM), a solid-state disk, and a software distribution medium.
The foregoing is a preferred embodiment of the present invention, and it will be apparent to those skilled in the art that various modifications may be made without departing from the principles of the invention, and such modifications are to be considered as within the scope of the invention.
Claims (8)
1. An image copying detection method based on a target is characterized by comprising the following steps:
acquiring an image to be detected, and performing copying detection on the image to be detected by adopting a preset first image copying detection model;
according to the first image reproduction detection model, clipping the image to be detected which is predicted to be non-reproduction according to a preset image preprocessing method to obtain a new image;
and performing feature extraction on the preprocessed new image by using a preset image feature extraction method, inputting the extracted features into a preset second image reproduction detection binary model to perform further image reproduction detection on the image to be detected, and finally determining whether the target in the image is reproduced or not according to the prediction result.
2. The first image duplication detection model of claim 1, wherein the categories are respectively:
(1) an image that does not contain a target; (2) the target is a copied image; (3) the target is a non-reproduced image;
wherein the target can be a type of target or a plurality of types of targets;
the method is also characterized in that the first image reproduction detection model has higher accuracy in predicting images containing targets with poor reproduction quality (such as images containing obvious screen frames or clear moire);
the first image capture detection model of claim 1, wherein the classification model is a convolutional neural network-based model, and wherein the model construction step comprises:
acquiring a first image sample set to be trained: wherein the sample set to be trained comprises three classes; dividing the first sample image to be trained into a training set, a verification set and a test set according to the proportion;
training the first image sample to be trained by adopting a convolutional neural network model to obtain a first image reproduction detection model;
the image acquisition of the first image sample set to be trained is characterized in that,
when the collected target is a copied image, a plurality of images of which the targets are copied are required to be shot by different shooting devices on different screens, wherein the copied image has not particularly high quality and has clear Moire patterns or obvious screen frames; partial antagonistic images are acquired for other two types of images, so that the model prediction precision is improved;
the obtained first image reproduction detection model is further characterized in that,
the probability that the copied image of the target with clear moire or obvious screen frame is predicted to be correct is higher for the relatively poor copying quality; and for a target copied image with very high copying quality and no frame or screen frame at the edge of the image, it may be predicted as a target non-copied image.
3. The pre-set image pre-processing method of claim 1,
when the predicted result obtained after the image to be detected is detected by the first image copying detection model is a non-copied image, the image to be detected is cut, and the edge of the image is removed, so that the situation that the edge of a mobile phone frame has little influence on the subsequent further copying detection result is prevented, and the preprocessed new image is obtained;
the preset image feature extraction method according to claim 1,
and carrying out graying on the new image obtained by the preprocessing, and then extracting image features such as binaryzation or color statistics.
4. The second image rendering detection classification model of claim 1,
the second image reproduction detection second classification model mainly detects images with higher reproduction quality and without clear moire or screen borders, and is characterized in that,
acquiring a second image sample set to be trained;
extracting image features from the images in the second sample set to be trained by using a preset image feature extraction method to obtain an image feature data set;
training the image characteristic data set by adopting a Support Vector Machine (SVM) to obtain a second image reproduction detection binary classification model;
the second image sample set to be trained is characterized in that,
the second image sample set to be trained comprises a plurality of original images, wherein the original images comprise two types, one type is a natural image containing a target and the other type is a non-reproduction image containing the target;
the natural image and the copied image are clear images, wherein the copied image does not contain the frame of a mobile phone or a computer and other equipment of the copied image and does not contain obvious moire fringes.
5. A device of an image reproduction detection method based on a target is characterized in that,
the method comprises the following steps: the system comprises a first image reproduction identification module, an image preprocessing module and a second image reproduction identification module;
the first image reproduction identification module is used for detecting whether an image to be detected contains a target or not and whether the target is a screen reproduction image or not, and the identification module can accurately identify the target reproduction image containing obvious moire or information such as a screen frame and the like; if no target is detected, directly exiting; if the fact that the target in the image is copied and has obvious frames or moire fringes is detected, outputting the image as the copied image; otherwise, temporarily judging that the image is a non-reproduction image containing the target;
the image preprocessing module is used for cutting the non-copied image with the target output result in the first image copying and identifying module, cutting off a small part of the edge of the image, and preventing the influence of a possibly existing few screen frames on the copying and detecting result of the second image;
the second image reproduction identification module is used for carrying out further reproduction identification on the original image which is predicted to be a non-reproduction image containing the target by the first image reproduction identification module; and finally, determining whether the target in the image to be detected is subjected to copying.
6. The apparatus of claim 5, further comprising a first image duplication detection model building module, wherein the model building module comprises a first image sample set acquisition subunit, a sample set division subunit, and a model training subunit;
the image first sample set obtaining subunit is used for obtaining three types of training images;
the image first sample set dividing subunit is configured to perform 8: 1: 1, wherein 80% of the first image reproduction detection model training, 10% of the first image reproduction detection model verification and 10% of the first image reproduction detection model testing are used;
and the model training subunit is used for inputting the image to be trained into a convolutional neural network model for training to obtain the first image copying and identifying model.
7. The apparatus of claim 5, further comprising a second image duplication detection model classification model building module, wherein the second classification model building module comprises a second image sample set acquisition subunit, an image feature extraction unit, and a model training subunit;
the second image sample set obtaining subunit is used for obtaining two types of images;
the image characteristic extraction unit is used for manually extracting the related image statistical characteristics of the grayed images in the second image sample set;
and the model training subunit performs model training on the extracted image feature set by using a Support Vector Machine (SVM) to obtain the second image reproduction detection model.
8. A medium comprising a stored computer program, wherein the computer program when executed implements the object-based image duplication detection method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111253036.7A CN113920434A (en) | 2021-10-27 | 2021-10-27 | Image reproduction detection method, device and medium based on target |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111253036.7A CN113920434A (en) | 2021-10-27 | 2021-10-27 | Image reproduction detection method, device and medium based on target |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113920434A true CN113920434A (en) | 2022-01-11 |
Family
ID=79243198
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111253036.7A Pending CN113920434A (en) | 2021-10-27 | 2021-10-27 | Image reproduction detection method, device and medium based on target |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113920434A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998086A (en) * | 2022-07-28 | 2022-09-02 | 合肥高维数据技术有限公司 | Method for manufacturing test sample of screen invisible watermark embedding program and test method |
| CN116168038A (en) * | 2023-04-26 | 2023-05-26 | 创新奇智(青岛)科技有限公司 | Image reproduction detection method and device, electronic equipment and storage medium |
-
2021
- 2021-10-27 CN CN202111253036.7A patent/CN113920434A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998086A (en) * | 2022-07-28 | 2022-09-02 | 合肥高维数据技术有限公司 | Method for manufacturing test sample of screen invisible watermark embedding program and test method |
| CN114998086B (en) * | 2022-07-28 | 2022-10-21 | 合肥高维数据技术有限公司 | Method for manufacturing test sample of screen invisible watermark embedding program and test method |
| CN116168038A (en) * | 2023-04-26 | 2023-05-26 | 创新奇智(青岛)科技有限公司 | Image reproduction detection method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Scherhag et al. | Detection of face morphing attacks based on PRNU analysis | |
| CN110598710B (en) | Certificate identification method and device | |
| CN112686812B (en) | Bank card inclination correction detection method and device, readable storage medium and terminal | |
| JP4772839B2 (en) | Image identification method and imaging apparatus | |
| US9740965B2 (en) | Information processing apparatus and control method thereof | |
| US8818112B2 (en) | Methods and apparatus to perform image classification based on pseudorandom features | |
| Qu et al. | Detect digital image splicing with visual cues | |
| US10043071B1 (en) | Automated document classification | |
| CN111325717B (en) | Mobile phone defect position identification method and equipment | |
| CN111311556B (en) | Mobile phone defect position identification method and equipment | |
| CN108805116A (en) | Image text detection method and its system | |
| CN113920434A (en) | Image reproduction detection method, device and medium based on target | |
| Do et al. | Automatic license plate recognition using mobile device | |
| US20150146991A1 (en) | Image processing apparatus and image processing method of identifying object in image | |
| CN110570442A (en) | Contour detection method under complex background, terminal device and storage medium | |
| CN111126393A (en) | Vehicle appearance refitting judgment method and device, computer equipment and storage medium | |
| CN110689003A (en) | Low-illumination imaging license plate recognition method and system, computer equipment and storage medium | |
| CN114445843A (en) | Card image character recognition method and device of fixed format | |
| CN113269752A (en) | Image detection method, device terminal equipment and storage medium | |
| JP2013250796A (en) | Number plate detection device, number plate detection method, and computer program | |
| EP2784721A2 (en) | Object detection apparatus | |
| Jaiswal et al. | Saliency based automatic image cropping using support vector machine classifier | |
| CN107886102B (en) | Adaboost classifier training method and system | |
| Long et al. | An Efficient Method For Dark License Plate Detection | |
| US11872832B2 (en) | Texture-based authentication of digital identity documents |
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 |