节点文献
基于几何/物理方法的图像伪造检测技术
Detecting Photo Forensics through Geometrical and Physical Clues
【作者】 张炜;
【导师】 操晓春;
【作者基本信息】 天津大学 , 计算机应用技术, 2010, 硕士
【摘要】 随着数字图像技术的发展和图像处理工具的逐渐强大,越来越多的伪造图像进入我们的视野,对信息的真实性提出挑战。如何利用图像本身的数据信息鉴别图像的真伪已经成为一个刻不容缓的任务。本文针对图像伪造的实际问题,以图像的透视几何和光学约束为研究重点,提出一套鉴别图像真实性的框架。此框架包含适用于数字图像复制检测的新图像特征描述及其匹配算法,对图像复制伪造进行不需要预先加入水印的盲检测,判断图像的真实性和有效性,对图像证据的可信度做出评价,辅助新闻,军事,法律,经济等决策。本论文包含一套新的基于图像内容的物理,几何信息的图像伪造盲检测算法。基于几何信息的盲检测算法采用极几何约束和平面单映性约束估计图像的伪造区域,针对媒体内容经过复杂变换后基于区域相似度的传统复制检测方法出现的高误报率问题,研究媒体内容变换前后满足平面单映矩阵的情况,包括相机旋转、缩放及内容共面等。在此一一对应的强约束不满足时,将多视几何学中的极几何约束引入图像复制检测领域,提出基于该弱约束的新约束及检测算法;基于物体阴影信息的盲检测算法采用Shadow Matte一致性和光源一致性约束,考虑从真实照片或视频帧中提取复杂光照模型尚不成熟的现状,采用基于简单光照一致性的复制检测,将计算机图形学中的Shadow Matte概念引入图像复制检测领域,提出新的基于光照一致性约束的检测算法。
【Abstract】 With the development of the digital image and the corresponding image processing tools, more and more multimedia contents make their way into our life and scope, which challenges the authenticity of the multimedia information. How to judge the authenticity of a given image with only the help of the image itself has become a emergency task.This paper aims to find a solution to the above problem, using the perspective image theory and optics information. We propose a new framework to detect forensics of the image. The framework includes copy detection method using feature description and matching algorithm. The proposed blind method, which does not require inserting additional watermark into the target image at the time recording, can be used to judge the authenticity of the image and help making decisions for news, military, laws and economics. We proposed a new set of methods based on geometrical and physical constraints. For the methods based on geometrical clue, we use the epipolar constraint and planar homography constraint. For the methods based on physical clue, we use the shadow matte and lighting source consistence. Considering extracting complex lighting model is difficult for now, shadow matte in graphics is introduced into image copy detection, and the simple lighting source consistence is also proposed as a physical constraint.
【Key words】 Image Forensic; Blind Detection; Planar Homography Constraint; Shadow Matte Consistence; Lighting Source Consistence; Multiple View Geometry;