WO2020119618A1 - 一种基于纹理特征融合的图像修复检测方法 - Google Patents
一种基于纹理特征融合的图像修复检测方法 Download PDFInfo
- Publication number
- WO2020119618A1 WO2020119618A1 PCT/CN2019/123943 CN2019123943W WO2020119618A1 WO 2020119618 A1 WO2020119618 A1 WO 2020119618A1 CN 2019123943 W CN2019123943 W CN 2019123943W WO 2020119618 A1 WO2020119618 A1 WO 2020119618A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- pixel
- features
- detection method
- feature
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
Definitions
- the invention relates to the field of image processing, in particular to an image repair detection method.
- Image repair refers to the process of reconstructing the lost or damaged part of the image.
- complex algorithms are used to reconstruct the lost or damaged image data.
- the true vision of the image is restored.
- repair movies restore degraded old photos, and remove watermarks, dates, and other information from pictures.
- some people with special intentions use the existing repair algorithm to maliciously tamper with the image, such as removing objects or people that exist in the image themselves, in an attempt to change the objective facts reflected in the original image. Therefore, how to effectively distinguish the authenticity of digital images becomes particularly important.
- Image repair algorithms mainly include two categories: the first category is diffusion-based image repair algorithms, which are mainly used for the removal of smaller objects, and will not leave perceptible repair traces in small areas; the second category is based on examples
- the image repair algorithm is mostly used to remove and repair larger objects in the image.
- the area to be tampered is particularly small, but most of the existing image repair detection algorithms are for the detection of the second type of algorithm. For small-sized image blocks, the detection results are not ideal.
- the existing detection technology is not very robust to JPEG compression, and JPEG compression will seriously affect the detection effect.
- the present invention proposes an image repair detection algorithm based on image texture feature fusion, which can realize repair detection and positioning of a small area of the image.
- An image repair detection method based on texture feature fusion includes the following steps;
- step (3) Perform the pixel-by-pixel texture feature extraction of step (2) again for the amount of change described in step (3);
- a local binary mode namely LBP
- a local phase quantization namely LPQ
- two local texture analysis operators are used to extract image texture features.
- the LBP feature is a uniform LBP feature with rotation-invariant properties.
- g c represents the central pixel value
- g a represents the neighborhood pixel value in the area of radius R
- P is the number of pixels in the neighborhood
- a 0, 1, ..., P-1
- the superscript riu2 represents a uniform LBP feature with rotation-invariant properties.
- N is an odd number greater than or equal to 3
- N 3.
- the change amount I'n (i, j) of the Tulaplas of the image in the image illuminance direction in step (3) is calculated by the following formula:
- n represents the number of repairs
- (i, j) represents the pixel coordinates
- L n (i, j) is the information that needs to be propagated, Indicates the direction of information dissemination, It is a measure of the change in the information to be disseminated.
- the fusion in step (5) is to connect the features of step (2) and step (4) in series to form the fusion feature.
- dimensionality reduction processing is also performed on the fusion feature.
- the classifier is an integrated classifier.
- removing abnormal points in step (7) includes excluding abnormally exposed points, and removing small false alarm areas according to morphological filtering.
- the present invention has the advantage that the present invention uses texture features to detect image repair, and the detection effect is better than conventional detection methods, especially for the repair of smaller-sized image blocks.
- the repair detection of JPEG compressed images is also robust.
- FIG. 1 is a flowchart of an image repair detection method of the present invention
- the image repair detection method based on texture feature fusion of the present invention mainly includes the following steps:
- S1 turn to grayscale image.
- the LBP feature is extracted pixel by pixel for the grayscale image. For each pixel, P pixels in the neighborhood of radius R are taken to extract uniform LBP features with rotation invariance.
- the calculation formula is as follows:
- g c represents the center pixel value
- g a represents the pixel value of the neighborhood in the area of radius R
- P is the number of pixels in the neighborhood
- riu2 represents the uniform LBP feature with rotation-invariant properties.
- g b Re(g b )+Im(g b )i is the bth of the Fourier transform vector
- the quantized value of the real part is usually ranked in the first four digits, and the corresponding imaginary value of the imaginary part is ranked in the last four digits.
- the resulting binary number is converted into a decimal number to obtain the LPQ feature.
- the LBP and LPQ features are calculated by mirror expansion.
- n represents the number of repairs
- (i, j) represents the pixel coordinates
- L n (i, j) is the information that needs to be propagated, Indicates the direction of information dissemination, It is a measure of the change in the information to be disseminated.
- L n (i, j) can be used Calculation.
- the subscripts xx and yy indicate the second-order partial derivatives of the x-axis and y-axis directions
- the gradient of Tulaplas can be used To represent.
- the vertical direction of the gradient is the direction of propagation use Indicates that it is obtained by rotating the gradient by 90 degrees.
- the LBP and LPQ features calculated in S2 and the LBP and LPQ features calculated in S4 are concatenated to obtain high-dimensional fusion features, which can better characterize the image of the repaired area and the unrepaired area.
- the fused features can also be dimension-reduced.
- the dimensionality reduction process can use the principal component analysis (PCA) method, or other dimensionality reduction methods that can retain the main information of the features.
- PCA principal component analysis
- the integrated classifier uses the trained model to judge each pixel one by one, returns +1 as the repair point, returns -1 as the unrepaired point, and finally excludes the abnormally exposed points in the image and removes the small ones according to the morphological filtering In case of false alarm, the final test result can be obtained.
- the point of excluding abnormal exposure refers to the partially underexposed and overexposed pixels in the image. Since the Tulaplas of these pixels is close to 0, the change of Tulaplas will also be very small, which has Certainly affected.
- the center point of the current window is considered to be an abnormally exposed area in the image, and it is determined as an unrepaired point.
- morphological filtering it refers to removing small false alarm areas through corrosion operation, expanding the repaired area detected through expansion operation, reducing false positives and improving true positives.
- a AUC F1-score Original image 0.8594 0.6811 QF 90 0.7471 0.6627
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
一种基于纹理特征融合的图像修复检测方法,包括以下步骤:将待检测图像转为灰度图;逐像素提取图像纹理特征;计算所述灰度图的图拉普拉斯在图像照度方向上的变化量;对所述变化量再次逐像素提取纹理特征;融合所有提取的纹理特征,形成融合特征;将所述融合特征输入分类器进行判决;移除异常点,得到检测结果。利用纹理特征实现对图像修复的检测,检测效果优于常规检测方法,尤其针对较小尺寸的图像块修复具有较好的检测效果,并且对JPEG压缩图像的修复检测也具有较好的鲁棒性。
Description
本发明涉及图像处理领域,尤其涉及图像修复的检测方法。
图像修复是指重建图像中丢失或损坏部分的过程,一般采用复杂的算法来重构已丢失或损坏的图像数据,在维持原图整体和谐统一的基础上,恢复了图像的真实视觉。一般用来修复电影、还原退化的老旧照片以及去除图片中的水印、日期等信息。但有些别具用心的人利用现有的修复算法对图像进行恶意篡改,例如移除图像中本身存在的物体或人物,企图改变原始图像所反映的客观事实。因此,如何有效区分数字图像的真实性,就变得尤为重要。
图像修复算法主要包括两大类:第一类是基于扩散的图像修复算法,其主要用于较小物体的移除,在小区域内不会留下可感知的修复痕迹;第二类是基于范例的图像修复算法,多用于图像中较大物体的移除修复。在一些篡改应用中,要篡改的区域特别小,但现有的图像修复检测算法,大多数是针对第二类算法的检测,对较小尺寸的图像块修复,检测结果并不理想。而且现有检测技术对JPEG压缩没有很好的鲁棒性,JPEG压缩会严重影响检测效果。
发明内容
有鉴于此,本发明提出一种基于图像纹理特征融合的图像修复检测算法,可实现对图像较小区域的修复检测定位。
一种基于纹理特征融合的图像修复检测方法,包括以下步骤;
(1)将待检测图像转为灰度图;
(2)逐像素提取图像纹理特征;
(3)计算所述灰度图的图拉普拉斯在图像照度方向上的变化量;
(4)对步骤(3)中所述变化量再次进行步骤(2)的逐像素提取纹理特征;
(5)融合步骤(2)、(4)提取的纹理特征,形成融合特征;
(6)将所述融合特征输入分类器进行判决;
(7)移除异常点,根据形态学滤波移除小的误警区域,得到检测结果。
优选地,步骤(2)中采用局部二值模式,即LBP,和局部相位量化,即LPQ,两种局部纹理分析算子提取图像纹理特征。
优选地,所述LBP特征为均匀的具有旋转不变性质的LBP特征。
对任一像素提取LBP特征的步骤为:
选取该像素作为中心像素,取半径为R的邻域中P个像素,提取均匀的具有旋转不变性质的LBP特征,计算公式为:
优选地,取P=16,R=2,得到1×18维LBP特征值。
对任一像素提取LPQ特征的步骤为:
以该像素为中心,取N×N像素的局部窗口,N为大于等于3的奇数,N=3,5,……11,在窗口内对像素值做离散傅里叶变换,对离散傅里叶变换向量的第b个成分,b=1、2、3、4,gb=Re(gb)+Im(gb)i的实部Re(gb)和虚部Im(gb)分别进行量化,得到量化返回值qk:
由此得到的8位二进制系数,将其转化为十进制数即得LPQ特征:
优选地,取N=3。
优选地,步骤(3)中图像的图拉普拉斯在图像照度方向上的变化量I′
n(i,j)由下式计算:
优选地,步骤(5)中的融合为将步骤(2)和步骤(4)的特征串联起来,形成所述融合特征。
优选地,形成融合特征后,还对所述融合特征进行降维处理。
优选地,所述分类器为集成分类器。
优选地,步骤(7)中移除异常点包括排除异常暴露的点,并根据形态学滤波移除小的误警区域。
相对于现有技术,本发明的优点在于:本发明利用纹理特征实现对图像修复的检测,检测效果优于常规检测方法,尤其针对较小尺寸的图像块修复具有较好的检测效果,并且对JPEG压缩图像的修复检测也具有较好的鲁棒性。
图1为本发明的图像修复检测方法流程图
如图1所示,本发明的基于纹理特征融合的图像修复检测方法,主要包括以下步骤:
S1、转灰度图。输入待检测彩色图像进行灰度转换,得到灰度图。
S2、逐像素地提取局部二值模式(Local binary pattern,LBP),局部相位量化(Local phase quantization,LPQ)特征。
首先对灰度图逐像素提取LBP特征,对于每个像素,取半径为R的邻域中P个像素,提取均匀的具有旋转不变性质的LBP特征,计算公式如下:
其中g
c表示中心像素值,
g
a表示半径为R的区域内的邻域像素值,P是邻域的像素点数,上角标riu2表示均匀的具有旋转不变性质的LBP特征。在本实施例中,取P=16,R=2,由此得到1×18维LBP特征。
然后对灰度图像逐像素提取LPQ特征,在取N×N像素的局部窗口,N为大于等于3的奇数,N=3,5,……11,通过对傅里叶变换的相位谱做去相关和均匀量化,得到图像像素的LPQ特征。在LPQ特征计算的时候,只考虑2维离散傅里叶变换的4个复数系数,其中,假设g
b=Re(g
b)+Im(g
b)i是傅里叶变换向量的第b个成分,b=1、2、3、4,将它的实部Re(g
b)和虚部Im(g
b)用一个简单的标量量化:
由此可得到8个二进制系数,这8个二进制系数按照一定的顺序排列即可得到LPQ特征,
排序时通常将实部的量化值排在前四位,对应的虚部量化值排在后四位,将所得二进制数编码转化为十进制数即得LPQ特征。
在本实施例中,取N=3。
对于图像的边缘像素,采用镜像扩展的方式计算LBP和LPQ特征。
S3、计算灰度图像的图拉普拉斯在图像照度方向上的变化量。图像的图拉普拉斯在图像照度方向上的变化量I′
n(i,j)实际上是图拉普拉斯在梯度垂直方向上的投影,可以由下式计算
其中n表示修复次数,(i,j)表示像素坐标,L
n(i,j)是需要传播的信息,
表示信息传播的方向,
是对要传播信息变化的一个度量。其中图像的图拉普拉斯L
n(i,j)可以用
计算。下标xx,yy分别表示对x轴和y轴方向求二阶偏导,
在连续数学中可以用图拉普拉斯的梯度
来表示。梯度的垂直方向即为传播的方向
用
表示,其将梯度的方向旋转90度得到。最终计算出I′
n(i,j)与灰度图的尺寸大小是相同的。在修复区域的I′
n(i,j)值接近于0,而在未修复区域I′
n(i,j)值偏大。
S4、对S3中得到的变化量提取LBP和LPQ特征。针对S3中提取的变化量I′
n(i,j),逐个像素提取LBP和LPQ特征。此过程的计算方法与S2中相同,只是针对图拉普拉斯在梯度垂直方向上的变化量提取纹理特征。
S5、特征融合。将S2计算的LBP、LPQ特征与S4中计算的LBP、LPQ特征串联起来,即得到高维的融合特征,可以更好地表征修复区域与未被修复区域的图像。为了提高计算速度,也可以将融合特征进行降维处理。降维处理可采用主成 分分析(PCA)方法,或其他能保留特征主要信息的降维方法。
S6、将融合特征输入到集成分类器测试。集成分类器(Ensemble classifier)先用来训练,将大量的已知修复与否的像素融合特征作为训练样本,输入到分类器训练,训练出一个判决较准确的模型。然后将测试样本,即测试图像每个像素的融合特征输入到训练好的模型,由模型做出判决,逐一判断每个像素是否经过修复。
S7、输出判决结果。集成分类器采用训练好的模型对每个像素逐一判决,返回+1的为修复点,返回-1的是未修复的点,最后排除图像中异常暴露的点并根据形态学滤波移除小的误警区域,即可得到最终的检测结果。排除异常暴露的点指的是针对图像中部分曝光不足和过度曝光的像素,由于这些像素的图拉普拉斯接近于0,导致图拉普拉斯的变化也会很小,对检测结果有一定影响。因此,当3*3窗口中像素值之和小于10或大于245时,则认为当前窗口的中心点是图像中异常曝光的区域,将其判定为未经修复的点。根据形态学滤波指的是通过腐蚀操作移除小的误警区域,通过膨胀操作扩大检测的修复区域,降低假阳性,提高真阳性。
可行性验证
在UCID图像数据库中验证。首先利用GIMP图像处理软件可以得到基于扩散的修复图像,包括各种不同的修复尺寸(8×8,32×32)以及对不同质量因子(QF=90,QF=75)的JPEG压缩图像。随机选取数据库中的338张图像用于训练,剩余的1000张图像用于测试,每张图像取50个正样本点,即经过修复的点和50个负样本点,即未被修复的点。采用ROC曲线的线下面积(Area under the curve,AUC)和F1-socre作为判别标准,检测结果如下表所示:
| AUC | F1-score | |
| 8×8 | 0.8594 | 0.6811 |
| 32×32 | 0.9025 | 0.7552 |
表1、对不同尺寸修复的检测结果
| AUC | F1-score | |
| 原图 | 0.8594 | 0.6811 |
| QF=90 | 0.7471 | 0.6627 |
| QF=75 | 0.6938 | 0.6171 |
表2、对不同质量因子(修复块大小为8×8时)图像修复的检测结果基于以上分析,本发明提出的方案在技术上是可行的。
所述领域的普通技术人员应当理解,以上所述仅为本发明的具体实施例而已,并不用于限制本发明,凡在本发明的精神和原则之内,所做的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。
Claims (10)
- 一种基于纹理特征融合的图像修复检测方法,其特征在于,包括以下步骤;(1)将待检测图像转为灰度图;(2)逐像素提取图像纹理特征;(3)计算所述灰度图的图拉普拉斯在图像照度方向上的变化量;(4)对步骤(3)中所述变化量再次进行步骤(2)的逐像素提取纹理特征;(5)融合步骤(2)、(4)提取的纹理特征,形成融合特征;(6)将所述融合特征输入分类器进行判决;(7)移除异常点,得到检测结果。
- 根据权利要求1所述的图像修复检测方法,其特征在于,步骤(2)中采用局部二值模式,即LBP,和局部相位量化,即LPQ,两种局部纹理分析算子提取图像纹理特征。
- 根据权利要求2所述的图像修复检测方法,其特征在于,所述LBP特征为均匀的具有旋转不变性质的LBP特征。
- 根据权利要求3所述的图像修复检测方法,其特征在于,对任一像素提取LBP特征的步骤为:选取该像素作为中心像素,取半径为R的邻域中P个像素,提取均匀的具有旋转不变性质的LBP特征,计算公式为:对任一像素提取LPQ特征的步骤为:以该像素为中心,取N×N像素的局部窗口,N为大于等于3的奇数,N=3,5,……11,在窗口内对像素值做离散傅里叶变换,对离散傅里叶变换向量的第b个成分,b=1、2、3、4,g b=Re(g b)+Im(g b)i的实部Re(g b)和虚部Im(g b)分别进行量化,得到量化返回值q k:
- 根据权利要求4所述的图像修复检测方法,其特征在于,在计算LBP特征值时取P=16,R=2,得到1×18维LBP特征值;在计算LPQ特征值时取N=3。
- 根据权利要求1-6任一项所述的图像修复检测方法,其特征在于,步骤(5)中的融合为将步骤(2)和步骤(4)的特征串联起来,形成所述融合特征。
- 根据权利要求7所述的图像修复检测方法,其特征在于,形成融合特征后,还对所述融合特征进行降维处理。
- 根据权利要求1-8任一项所述的图像修复检测方法,其特征在于,所述分类器为集成分类器。
- 根据权利要求1-9任一项所述的图像修复检测方法,其特征在于,步骤(7)中移除异常点即排除异常暴露的点,并根据形态学滤波移除小的误警区域。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811523591.5A CN109785286B (zh) | 2018-12-12 | 2018-12-12 | 一种基于纹理特征融合的图像修复检测方法 |
| CN201811523591.5 | 2018-12-12 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020119618A1 true WO2020119618A1 (zh) | 2020-06-18 |
Family
ID=66495894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/123943 Ceased WO2020119618A1 (zh) | 2018-12-12 | 2019-12-09 | 一种基于纹理特征融合的图像修复检测方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109785286B (zh) |
| WO (1) | WO2020119618A1 (zh) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785286B (zh) * | 2018-12-12 | 2021-04-30 | 中国科学院深圳先进技术研究院 | 一种基于纹理特征融合的图像修复检测方法 |
| CN112329781B (zh) * | 2020-11-04 | 2021-05-11 | 哈尔滨市科佳通用机电股份有限公司 | 基于图像修复的货车制动梁支柱拉铆销套环丢失检测方法 |
| CN113222843B (zh) * | 2021-05-10 | 2023-11-10 | 北京有竹居网络技术有限公司 | 一种图像修复方法及其相关设备 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100091127A1 (en) * | 2008-09-30 | 2010-04-15 | University Of Victoria Innovation And Development Corporation | Image reconstruction method for a gradient camera |
| CN104463853A (zh) * | 2014-11-22 | 2015-03-25 | 四川大学 | 一种基于图像分割的阴影检测与去除算法 |
| CN105893649A (zh) * | 2015-03-23 | 2016-08-24 | 温州大学 | 一种基于最优化模型的交互式图像重新着色方法 |
| CN109785286A (zh) * | 2018-12-12 | 2019-05-21 | 中国科学院深圳先进技术研究院 | 一种基于纹理特征融合的图像修复检测方法 |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573707B (zh) * | 2014-12-17 | 2017-11-14 | 安徽清新互联信息科技有限公司 | 一种基于多特征融合的车牌汉字识别方法 |
| CN105117688B (zh) * | 2015-07-29 | 2018-08-28 | 重庆电子工程职业学院 | 基于纹理特征融合和svm的人脸识别方法 |
| CN106960435A (zh) * | 2017-03-15 | 2017-07-18 | 华中师范大学 | 一种jpeg图像双压缩自动检测方法 |
| CN107862709B (zh) * | 2017-09-28 | 2020-03-27 | 北京华航无线电测量研究所 | 一种多方向模式连接规则的图像纹理描述方法 |
| CN108022197B (zh) * | 2017-11-21 | 2021-03-12 | 安徽大学 | 一种基于多纹理度划分的医学图像可逆信息隐藏方法 |
-
2018
- 2018-12-12 CN CN201811523591.5A patent/CN109785286B/zh active Active
-
2019
- 2019-12-09 WO PCT/CN2019/123943 patent/WO2020119618A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100091127A1 (en) * | 2008-09-30 | 2010-04-15 | University Of Victoria Innovation And Development Corporation | Image reconstruction method for a gradient camera |
| CN104463853A (zh) * | 2014-11-22 | 2015-03-25 | 四川大学 | 一种基于图像分割的阴影检测与去除算法 |
| CN105893649A (zh) * | 2015-03-23 | 2016-08-24 | 温州大学 | 一种基于最优化模型的交互式图像重新着色方法 |
| CN109785286A (zh) * | 2018-12-12 | 2019-05-21 | 中国科学院深圳先进技术研究院 | 一种基于纹理特征融合的图像修复检测方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109785286B (zh) | 2021-04-30 |
| CN109785286A (zh) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111047522B (zh) | 一种基于边缘生成的图像修复方法 | |
| Wu et al. | Busternet: Detecting copy-move image forgery with source/target localization | |
| CN113744153A (zh) | 双分支图像修复伪造检测方法、系统、设备及存储介质 | |
| CN111612741B (zh) | 一种基于失真识别的精确无参考图像质量评价方法 | |
| CN112069891A (zh) | 一种基于光照特征的深度伪造人脸鉴别方法 | |
| Choudhary et al. | Image forgery detection system using VGG16 UNET model | |
| CN110457996B (zh) | 基于vgg-11卷积神经网络的视频运动对象篡改取证方法 | |
| WO2017080196A1 (zh) | 基于人脸图像的视频分类方法和装置 | |
| WO2020119618A1 (zh) | 一种基于纹理特征融合的图像修复检测方法 | |
| CN110503606A (zh) | 一种提高人脸清晰度的方法 | |
| Hilles et al. | Latent fingerprint enhancement and segmentation technique based on hybrid edge adaptive DTV model | |
| CN108830829A (zh) | 联合多种边缘检测算子的无参考质量评价算法 | |
| CN110188828A (zh) | 一种基于虚拟样本集成学习的图像来源鉴别方法 | |
| Du et al. | Eyeglasses removal from facial images | |
| CN108681997A (zh) | 基于改进lbp特征的无参考多失真图像质量评价方法 | |
| CN108629771B (zh) | 一种具有尺度鲁棒性的图像质量盲评价方法 | |
| Sharma et al. | Spliced Image Classification and Tampered Region Localization Using Local Directional Pattern. | |
| CN114240795A (zh) | 一种融合偏色校正的图像去沙尘暴方法 | |
| CN106709915B (zh) | 一种图像重采样操作检测方法 | |
| CN110335283A (zh) | 图像修复方法、装置、设备以及计算机可读存储介质 | |
| Sharma et al. | Anti-forensic technique for median filtering using L 1-L 2 TV model | |
| Zhu et al. | Recaptured image detection through enhanced residual-based correlation coefficients | |
| CN111275687B (zh) | 一种基于连通区域标记的细粒度图像拼接检测方法 | |
| CN107798676B (zh) | 基于多尺度自然场景统计建立的无参考图像质量感知方法 | |
| Liu et al. | Recaptured image detection based on DCT coefficients |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19895305 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.11.2021) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19895305 Country of ref document: EP Kind code of ref document: A1 |

