Free Academic Seminars And Projects Reports

Full Version: Detection of Resampling Supplemented with Noise Inconsistencies Analysis for Image F
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Abstract

When two or more images are spliced together, to create high quality and consistent image forgeries, almost always geometric transformations such as scaling or rotation are needed. These procedures are typically based on a resampling and interpolation step. In this paper, we introduce a blind method capable of finding traces of resampling and interpolation. Unfortunately, the proposed method, as well as other existing interpolation/resampling detectors, is very sensitive to noise. The noise degradation causes that detectable periodic correlations brought into the signal by the interpolation process become corrupted and difficult to detect. Therefore, we also propose a novel method capable of dividing an investigated image into various partitions with homogenous noise levels. Adding locally random noise may cause inconsistencies in the image s noise. Hence, the detection of various noise levels in an image may signify tampering.

Introduction

Without a doubt, image authenticity is significant in many social areas and plays a crucial role in people s lives. In this paper we focus on blind digital image authentication [9, 10, 6, 16, 16, 8, 14, 5]. The blind approach is regarded as the new direction and is a burgeoning research field. In contrast to active approaches, passive approaches do not need any explicit prior information about the image. They work in the absence of any digital watermark or signature and are based on the image characteristic. The area of blind digital image authentication is growing rapidly and the results obtained in this dissertation, as well as results from other existing blind techniques, promise a significant improvement of forgery detection in the never ending game between image forgery creators and image forgery detectors.
When two or more images are spliced together , in order to create a consistent and high quality tampering, geometric transformations such as resizing, rotating or skewing are almost always needed. These procedures are typically based on a resampling and interpolation (nearest neighbor, linear, cubic, etc.) step. Despite the importance, massive usage1 and history2 of interpolation, to our knowledge, there exist only a few published works concerned with the specific and detectable statistical changes brought into the signal by this process. Therefore, in this paper we analytically show periodic properties present in the covariance structure of interpolated signals and their nth derivatives. Without the detailed knowledge of how the statistics of the signal is changed by the interpolation process, applications based on statistical approaches working with resampled/interpolated signals or with their derivatives can yield miscalculations and unexpected results. Furthermore, we briefly show a blind, efficient and automatic method capable of detecting the traces of resampling and interpolation. The method is based on a derivative operator and radon transformation.
Probably the main weakness of the mentioned interpolation/ resampling detector is its high sensitivity to noise. The noise degradation causes that detectable periodic correlations brought into the signal by the interpolation process become corrupted and difficult to detect. So, the mentioned weakness is common for all existing resampling detectors. Generally, additive noise is the main cause of failure of most existing blind authentication methods. These methods are able to work correctly only when the amount of present noise degradation is small. Based on these facts, in this paper we propose a novel method capable of dividing an in- vestigated image into various partitions with homogenous noise levels. Adding locally random noise may cause inconsistencies in the image s noise. Therefore, the detection of various noise levels in an image may signify tampering. We assume local additive white Gaussian noise.
The rest of the paper is organized as follows. The next section summarizes previous published papers concerned with the detection of scaling and rotation. After this, some basic notations and definitions are given to build up the necessary mathematical background. Section 4 analyzes and analytically shows hidden periodic properties present in interpolated signals. Section 5 introduces a method capable of detecting the traces of scaling and rotation. The following section proposes a novel method capable of segmenting an investigated image using the local noise level. Each step of the method is discussed in detail. Section 7 contains experiments to demonstrate the outcomes of the method. In section 8 important properties of the method and obtained results are discussed. The last section summarizes the work that has been done in this paper.