KR102195763B1 - Method and apparatus for processing image - Google Patents

Abstract

The image processing method performed by the disclosed image processing apparatus includes converting an image of a pixel domain into a DCT domain, dividing it into a plurality of patches, and then extracting a DCT histogram feature for each patch. And performing clustering divided into manipulation patches, and detecting the clustered manipulation patches as manipulation positions of the image.

Description

Image processing method and apparatus {METHOD AND APPARATUS FOR PROCESSING IMAGE}

The present invention relates to an image processing method and apparatus, and more particularly, to an image processing method and apparatus for determining a manipulation position with respect to an image suspected of manipulation.

Double JPEG technology (Mauro Barni, Luca Bondi, Nicol`o Bonettini, Paolo Bestagini, Andrea Costanzo, Marco Maggini, Benedetta) as a method of detecting the manipulated position for image files in JPEG format or image files in other formats but past JPEG format. Tondi, and Stefano Tubaro, "Aligned and non-aligned double jpeg detection using convolutional neural networks," Journal of Visual Communication and Image Representation, vol. 49, pp. 153-163, 2017.). Double JPEG technology is a technology that determines whether an image is compressed once or twice as a JPEG, and is a technology that detects manipulation by taking advantage of the fact that when a general JPEG image is edited and saved with an image editing program, a double JPEG trace remains. However, such double JPEG technology has a limitation in determining the manipulation location only after preparing data for double JPEG detection to determine the manipulation location and then learning with a double/single classifier.

In order to remove the limitations of the learning-based classifier according to this double JPEG technology, a method of determining the manipulation location only with images that are suspicious of manipulation (Davide Cozzolino and Luisa Verdoliva, "Singleimage splicing localization through autoencoder-based anomaly detection," in Information Forensics and Security. (WIFS), 2016 IEEE International Workshop on.IEEE, 2016, pp. 1-6.) can be considered. However, the method of determining the manipulated position only with an image that is suspicious of manipulation can determine the manipulated position only for manipulation through splicing.

According to an embodiment of the present invention, image files in JPEG format or image files in other formats but past JPEG format are manipulated through various manipulation actions including copy-move and splicing. It provides an image processing method and apparatus capable of detecting.

The problem to be solved of the present invention is not limited to those mentioned above, and another problem to be solved that is not mentioned will be clearly understood by those of ordinary skill in the art from the following description.

An image processing method performed by an image processing apparatus according to an aspect of the present invention includes the steps of converting an image of a pixel domain into a Discrete Cosine Transform (DCT) domain, dividing it into a plurality of patches, and extracting a DCT histogram feature for each patch; And performing clustering for dividing patches from which the DCT histogram feature is extracted into a manipulated patch and a non-manipulated patch, and detecting the manipulated patch as a manipulated position of the image.

Here, in the performing of the clustering, a Euclidean distance is used as a distance metric, and the K-means clustering technique is repeatedly performed a plurality of times to select clustering having the smallest distance.

The detecting as the manipulation position may indicate a normal state or an abnormal state of a result of the clustering.

The steady state or the abnormal state is based on the center value of each DCT histogram feature, the distance between the center value of the manipulated patch cluster and the center value of the unmanipulated patch cluster, and the center value and median distance value in each cluster. Can be represented by

An image processing apparatus according to another aspect of the present invention includes a feature extraction unit for converting an image of a pixel domain into a DCT domain, dividing it into a plurality of patches, and extracting a DCT histogram feature for each patch, and the patches from which the DCT histogram feature is extracted. It may include a cluster unit that performs clustering divided into a manipulated patch and a non-manipulated patch, and a location detection unit that detects the manipulated patch as a manipulated position of the image.

Here, the cluster unit may select clustering having the smallest distance by using Euclid distance as a distance metric and repeatedly performing the K-means clustering technique a plurality of times.

The location detector may indicate a normal state or an abnormal state of a result of the clustering.

The normal state or the abnormal state can be expressed based on the center value of each DCT histogram feature, the distance between the center value of the manipulated patch cluster and the center value of the unmanipulated patch cluster, and the center value and median distance value within each cluster. have.

According to an embodiment of the present invention, it is possible to detect the manipulation position of an image through various manipulation actions including pasting and splicing with respect to an image file in a JPEG format or an image file in another format but past JPEG format. .

1 is a configuration diagram of an image processing apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating an image processing method performed by an image processing apparatus according to an embodiment of the present invention.
3 is a state diagram showing a process of changing an image according to an image processing method according to an embodiment of the present invention.

Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various forms, and only these embodiments make the disclosure of the present invention complete, and those skilled in the art to which the present invention pertains. It is provided to fully inform the person of the scope of the invention, and the scope of the invention is only defined by the claims.

In describing the embodiments of the present invention, detailed descriptions of known functions or configurations will be omitted except when actually necessary in describing the embodiments of the present invention. In addition, terms to be described later are terms defined in consideration of functions in an embodiment of the present invention, which may vary according to the intention or custom of users or operators. Therefore, the definition should be made based on the contents throughout this specification.

1 is a configuration diagram of an image processing apparatus 100 according to an embodiment of the present invention.

As shown in FIG. 1, the image processing apparatus 100 according to an embodiment of the present invention may include a feature extraction unit 110, a cluster unit 120, and a location detection unit 130. For example, the image processing apparatus 100 and the feature extraction unit 110, the cluster unit 120, and the location detection unit 130 that may configure the same may be implemented as a Graphics Processing Unit (GPU).

The feature extraction unit 110 may convert an image of a pixel domain into a discrete cosine transform (DCT) domain, divide it into a plurality of patches, and extract a DCT histogram feature for each patch.

The cluster unit 120 may perform clustering in which patches from which the DCT histogram feature is extracted are divided into manipulated patches and unmanipulated patches. For example, the cluster unit 120 may use a Euclidean distance as a distance metric, and repeat the K-means clustering technique multiple times to select clustering having the smallest distance.

The location detection unit 130 may detect the manipulation location of the image through various manipulation actions including pasting or splicing of manipulation patches. In addition, the location detection unit 130 may identify a normal state or an abnormal state of a result of clustering by the cluster unit 120 so as to be identifiable. For example, the location detection unit 130 determines the normal or abnormal state of the result of clustering, the center value of each DCT histogram feature, the distance between the center value of the manipulated patch cluster and the center value of the unmanipulated patch cluster, within each cluster. It can be expressed based on the center value and the median distance value.

2 is a flowchart illustrating an image processing method performed by an image processing apparatus according to an embodiment of the present invention, and FIG. 3 is a flowchart illustrating a process of changing an image according to an image processing method according to an embodiment of the present invention. It is a state diagram.

Hereinafter, an image processing method performed by the image processing apparatus 100 according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 3.

First, an image file in a JPEG format or an image file in another format but passed through the JPEG format in the past is input to the feature extraction unit 110 of the image processing apparatus 100.

Then, the feature extraction unit 110 extracts the DCT histogram feature from the input image file.

As such, the reason for changing the pixel domain to the DCT domain is that JPEG traces remain relatively better in the DCT domain than in the pixel domain. However, it takes a long time if the image processing for extracting the DCT histogram is processed by a CPU (Central Processing Unit), which has a relatively smaller number of Arithmetic Logic Units (ALUs) than GPUs. Meanwhile, when the image processing apparatus 100 and the feature extraction unit 110 are implemented as GPUs, it is necessary to perform a pre-processing process in order to shorten the image processing time for extracting the DCT histogram.

Referring to this pre-processing process, the feature extraction unit 110 converts the input pixel domain image into a DCT domain and divides it into a plurality of patches. For example, the feature extraction unit 110 may convert the input image file into YCbCr and divide it into 64×64 patches using only the Y channel.

In addition, the feature extraction unit 110 performs a DCT coefficient D through a 2D convolution operation of the image I and the DCT basis functions _{c1 and c2} of the 8×8×64 block for each patch as shown in Equation 1 Extract _{c1 and c2} .

Thereafter, the feature extraction unit 110 uses a list of bin boundary values b of the histogram, and uses a cumulative density function (CDF) B _{c1,c2 for} each bin boundary as shown in Equation 2 Calculate

는 (D_c1,c2-b)의 값을 증폭하는데 사용된 이득이고, 이 값을 0 또는 1로 바꿔서 시그모이드(sigmoid) 함수를 계산한다.here,

Is the gain used to amplify the value of (D _c1,c2 -b) _{, and} converts this value to 0 or 1 to calculate the sigmoid function.

Next, the feature extraction unit 110 obtains the histogram features Z _{c1 and c2} by differentiating the cumulative distribution function by convolving [1, -1] for each bin of the histogram as shown in Equation (3).

Here, the empty range is set to [-5,5], and features of 10x63 dimensions per 64x64 patch of Y channel of the image file are extracted.

Next, the cluster unit 120 of the image processing apparatus 100 divides the patches from which the DCT histogram feature is extracted through step S210 into a manipulated patch and a non-manipulated patch, thereby performing clustering, thereby performing a JPEG format image file or other format. However, it is possible to determine the manipulation location for image files that passed the JPEG format in the past. For example, the cluster unit 120 may perform a K-means (KM) clustering technique in which data points are grouped into K groups, and may be divided into a manipulated patch cluster and an unmanipulated patch cluster. For example, the cluster unit 120 may use a Euclidean distance as a distance metric. In addition, the cluster unit 120 may repeatedly perform the K-means clustering technique a plurality of times to select clustering having the smallest distance. For example, the cluster unit 120 may repeatedly perform the K-means clustering technique five times to select clustering having the smallest distance in order to avoid local optima.

Accordingly, the location detection unit 130 of the image processing apparatus 100 detects the manipulation patch cluster clustered by the cluster unit 120 as a manipulation location through various manipulation actions including pasting and splicing. Meanwhile, a classification error may be included in the clustering result performed through step S220 with respect to the image suspected of manipulation. Accordingly, the location detection unit 130 of the image processing apparatus 100 may check whether the clustering result in step S220 is normal or abnormal, and display it to be identifiable.

When clustering is well performed by the clustering process in step S220, the distance between clusters is long, and when one class is clustered into two, the distance between clusters is short. Using this characteristic, the location detection unit 130 may represent the clustering result of step S220 as a metric T value using Equation 4.

Where med(·) is the median operator, d(x,y) is the squared Euclidean distance between x and y, f is the feature, P is the feature set of the unmanipulated patch cluster, and M is the manipulated patch cluster Is the feature set of, c _p is the center value of the unmanipulated cluster, c _m is the center value of the manipulated cluster, and c _all is the center value of all clusters.

As such, the location detection unit 130 determines the normal or abnormal state of the clustering result in step S220, the center value of each DCT histogram feature, the distance between the center value of the manipulated patch cluster and the center value of the unmanipulated patch cluster, and each cluster. It can be expressed based on the inner center value and the median distance value.

According to an embodiment of the present invention as described so far, after dividing an image in a JPEG format or an image in another format but passing through the past JPEG format into a plurality of patches, DCT histogram features are extracted, and the DCT histogram features are extracted. By performing clustering that is divided into a manipulated patch and a non-manipulated patch, it is possible to detect the manipulation position of the image through various manipulation actions including pasting and splicing.

Combinations of each block in the block diagram attached to the present invention and each step in the flowchart may be performed by computer program instructions. Since these computer program instructions can be mounted on a processor of a general purpose computer, special purpose computer, or other programmable data processing equipment, the instructions executed by the processor of the computer or other programmable data processing equipment are shown in each block or flow chart of the block diagram. Each step creates a means to perform the functions described. These computer program instructions can also be stored in computer-usable or computer-readable memory that can be directed to a computer or other programmable data processing equipment to implement a function in a particular way, so that the computer-usable or computer-readable memory It is also possible to produce an article of manufacture in which the instructions stored in the block diagram contain instruction means for performing the functions described in each block or flow chart. Computer program instructions can also be mounted on a computer or other programmable data processing equipment, so that a series of operating steps are performed on a computer or other programmable data processing equipment to create a computer-executable process to create a computer or other programmable data processing equipment. It is also possible for the instructions to perform the processing equipment to provide steps for performing the functions described in each block of the block diagram and each step of the flowchart.

In addition, each block or each step may represent a module, segment, or part of code comprising one or more executable instructions for executing the specified logical function(s). In addition, it should be noted that in some alternative embodiments, functions mentioned in blocks or steps may occur out of order. For example, two blocks or steps shown in succession may in fact be performed substantially simultaneously, or the blocks or steps may sometimes be performed in the reverse order depending on the corresponding function.

The above description is merely illustrative of the technical idea of the present invention, and those of ordinary skill in the art to which the present invention pertains will be able to make various modifications and variations without departing from the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, but to explain the technical idea, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

100: image processing unit
110: feature extraction unit
120: cluster unit
130: location detection unit

Claims (8)

An image processing method performed by an image processing device, comprising:
Converting an image in the pixel domain into a DCT (Discrete Cosine Transform) domain, dividing it into a plurality of patches, and extracting a DCT histogram feature for each patch;
Clustering the patches from which the DCT histogram feature is extracted into a manipulated patch and a non-manipulated patch,
Detecting the manipulation patch as a manipulation location of the image,
The detecting as the manipulation position may include indicating a normal state or an abnormal state of a result of the clustering, and indicating the normal state or the abnormal state, a center value of each DCT histogram feature, a center value of the manipulation patch cluster, and the Based on the distance of the center value of the unmanipulated patch cluster, the center value and the median distance value within each cluster
Image processing method. The method of claim 1,
In the performing of the clustering, a Euclidean distance is used as a distance metric, and the K-means clustering technique is repeatedly performed a plurality of times to select clustering having the smallest distance.
Image processing method. delete delete A feature extracting unit that converts an image in the pixel domain into a DCT (Discrete Cosine Transform) domain, divides it into a plurality of patches, and extracts DCT histogram features for each patch;
A cluster unit for performing clustering that divides the patches from which the DCT histogram feature is extracted into a manipulated patch and a non-manipulated patch,
A position detection unit for detecting the manipulation patch as a manipulation location of the image,
The location detection unit may indicate a normal state or an abnormal state of a result of the clustering, and determine the normal state or the abnormal state, a center value of each DCT histogram feature, a center value of the manipulated patch cluster, and the unmanipulated patch cluster. Based on the distance of the center value of, and the median distance value within each cluster
Image processing unit. The method of claim 5,
The cluster unit uses Euclidean distance as a distance metric, and selects clustering having the smallest distance by repeatedly performing the K-means clustering technique multiple times.
Image processing unit. delete delete

Images