基于机器视觉的小目标复制图像篡改检测方法  

Method for Tamper Detecting of Copied Images of Small Targets Based on Machine Vision

作　　者：赵宝水 黄海龙[1] 田昊 ZHAO Bao-shui;HUANG Hai-long;TIAN Hao(Liaoning University of Technology,Liaoning Jinzhou 121000,China)

机构地区：[1]辽宁工业大学,辽宁锦州121000

出　　处：《计算机仿真》2021年第8期227-230,304,共5页Computer Simulation

基　　金：辽宁省科技厅自然基金指导计划项目(JP2016019);辽宁省教育厅科学研究项目(JQL201915404)。

摘　　要：修图软件的多样化发展对篡改检测效果提出了新的挑战,针对小目标复制图像,以机器视觉作为技术支持,设计一种图像篡改检测方法。利用Prewitt方向导数近似算子,检测复制图像边缘,通过转换图像检测空间为参数空间,完成模板匹配与匹配值加权;经过非极大值抑制处理角点响应函数的各个方向,获取直角角点输出及其方向,扩充角点区域后明确检测区域,采用zernike矩离散化处理图像的实部与虚部,提取图像特征,结合图像各角点坐标与亚像素坐标,判定伪造区域与真实区域,实现图像篡改检测。从公开图像数据集CASIA中随机选取实验数据集,经过对比评估指标数据发现,在图像数据受到不同程度的图像压缩攻击和高斯噪声污染时,所提方法仍具有显著的稳定性与有效性,通过AUC值进一步验证了较好的综合性能与实际应用性。Based on machine vision,this paper designed a novel image tamper detection method.Prewitt directional derivative approximation operator was used to detect the edge of the copied image.Image detection space was transformed into parameter space to complete template matching and weighting of matching value.According to the non-maximum suppression,all directions of the corner response function were processed,and the output and direction of the right-angle corner were obtained.The detection area was determined after the corner area was expanded.Zernike moment separation was used to de scatter the real and imaginary parts of the image in order to extract image features.The coordinates of each corner and sub-pixel of the image were introduced to determine the forgery area and the real area,realizing the image tampering detection.The results show that the method is superior to the traditional method,and has significant stability and effectiveness.Additionally,the AUC value further verifies that the method has good comprehensive performance and applicability.

关 键 词：机器视觉 小目标 复制粘贴 图像篡改 篡改检测 图像特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]