期刊
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
卷 52, 期 -, 页码 24-32出版社
ACADEMIC PRESS INC ELSEVIER SCIENCE
DOI: 10.1016/j.jvcir.2018.01.015
关键词
Camera model identification; Classification; Digital forensics; Image features; Illumination dependency; Overfitting; Scene content
资金
- Media Asia Lab, Ministry of Electronics and Information Technology (MeitY), Govt. of India [PhD-MLA/4(13)/2015-16]
State-of-the-art techniques for Camera Model Identification operate by extracting different features from the training image set and incorporating those features to predict the source of test images using machine learning. Though the existing approaches perform efficiently for images captured in natural daylight or bright illumination conditions, the state-of-the-art lacks sufficient experiments and results to evaluate efficiency of such schemes for images captured in dark illumination conditions. In this paper, we present a set of experiments to assess the impact of illumination conditions, on image source classification problem, and also propose an image filtering based technique to eliminate the adverse effects of scene illumination on source classification accuracy. Our experimental results prove that the performance efficiency of existing feature based source classification techniques, is indeed dependent on the illumination conditions. The proposed strategy enables our source classification model to achieve high efficiency as compared to the state-of-the-art, under all illumination conditions.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据