Overview

This dissertation, Automatic Source Camera Identification by Lens Aberration and JPEG Compression Statistics by Kai-san, Choi, 蔡啟新, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of the thesis entitled Automatic Source Camera Identification by Lens Aberration and JPEG Compression Statistics Submitted by Choi Kai San for the degree of Master of Philosophy at The University of Hong Kong in December 2006 With the advent of high resolution digital cameras and sophisticated image edit- ing software, digital images can be manipulated easily. These manipulations can range from removing red eye effects to changing the background, and in some cases can significantly alter people's perception of the events recorded by the image. Usually, these alternations leave no observable traces. This undermines the traditional belief that seeing is believing, and creates a situation where im- ages no longer hold the status of definitive records of events. The credibility of digital images hinders their use for news reporting and crime investigation. As a result, there is a need to detect images that have been tampered with. In image forensics, one way to ascertain the authenticity of an image is by identifying its source camera.This study aims at identifying promising features for classifying images ac- cording to their source cameras. The problem can be approached by either exploiting the cameras' hardware defects or utilizing the differences of image processing algorithms between different cameras. By the first approach, we pro- posetouselensradialdistortion. Asalllensesinevitablypossessdifferentdegrees oflensradialdistortion, theyleaveuniquefootprintsontheimagesproduced. By examining every image's lens radial distortion, we can identify the source camera of each image. By the latter approach, we propose to use JPEG compression statistics. JPEG compression is one of the popular image formats in digital cam- eras. Sincedifferentcameramodelshavetheirowncustomizedfilesizeandimage quality tradeoff, JPEG compression leaves unique footprints on the images. We collect a number of statistics left behind by the JPEG compression for classifica- tion. Inordertoevaluatetheeffectivenessofourproposedfeaturesinsourcecamera identification, we extract parameters from lens radial distortion and JPEG com- pressionstatisticsfromeachimageandthenpasstheparameterstotrainandtest asupportvectormachineclassifier. Extensiveexperimentsareconductedandthe results show that both lens radial distortion and JPEG compression statistics are viable approaches with a high degree of accuracy. We further improve the efficiency of our proposed features by feature reduc- tion. To avoid the well-known curse of dimensionality problem which is causedby high dimension feature sets, a feature reduction method is required. We use stepwise discriminant analysis, together with cross validation analysis, to reduce the number of features. Simulations are carried out to compare the performance of full feature set, reduced feature set and randomly selected feature set. The results show that the reduced feature set decreases the processing time while also maintaining or even improving the classification accuracy under some circum- stances. An abstract of exactly 432 words DOI: 10.5353/th_b3890234 Subjects: JPEG (Image coding standard)Error analysis (Mathematics)Image compressionDigital camerasPhotographic lensesAberration

Full Product Details

9781361469156

Table of Contents

Reviews

Author Information

Tab Content 6

Countries Available