A Deep Learning Approach For Image Splicing Detection Using Error Level Analysis

Abstract

In this digital era, camera-equipped cell phones, scanners, and digital cameras are becoming progressively popular, thus making it simple to obtain digital images. These images are used as a medium for exchanging the information in social media, news, court, communication, etc. Digital images are often seen as an evidence of reality or a fact. Images can be easily and cheaply manipulated due to the wide availability of photo-editing softwares with the intent and purpose to benefit one party. Therefore, the image that is already manipulated can be used for fake news, evidence, and any publication or to gain popularity to mislead others. As a result, the authenticity of the digital image can no longer be taken for granted. It becomes very challenging for the end users to distinguish whether the image is original or altered. Therefore, image verification has become a significant problem in ensuring the validity of digital images with applications in areas such as medical, government, finance, law enforcement, etc. Earlier, machine learning algorithms were utilized for classification or regression problems. But, nowadays with the availability of huge data, classical machine learning methods are not working well. Therefore, to handle large and complex data, powerful methods and systems are required. The concept of deep learning is the right solution to this problem. Deep Learning models, with their multi-level structures, are very helpful in extracting complicated information from input images. This motivates to propose the technique that consists of deep learning models to detect image forgeries such as copy-move, splicing forgeries, etc. In copy-move forgery, a portion of an original image is copied and pasted within the same image while in image splicing, the copied portion is pasted on another image. It is difficult to identify the spliced images by the human visual system. Thus, there is a need to develop an efficient technique for the detection of spliced image. The proposed technique is designed for the detection of image splicing forgery by making use of Convolutional Neural Network (CNN). It is a deep learning neural network which learns patterns from given data. Firstly, preprocessing is performed on images. For preprocessing of images, Error Level Analysis (ELA) technique is utilized. It generates comparable error levels by examining compression artifacts in the given images. After processing with ELA tool, the images are resized and normalized. Then these preprocessed images are fed to CNN for the classification. For extraction of features, CNN uses convolutional layers and for classification, it uses fully connected layers. It extracts features from the preprocessed images automatically and then classifies them between two classes, i.e., authentic and forged images. The experimental results show that the performance of the proposed method, when compared to some existing methods, is better in terms of accuracy, precision, recall, and F1 score.