Efficient Framework for Digital Image Forgery Detection

Abstract

Forged images are used for some illegal activity like to generate fake news related to political, social, personal, medical, and legal, etc. Advanced technology makes it easy to develop a doctored image. It s on our fingertip. So need to develop such a system that can authenticate and help to make the right perception about original and fake images. Various post processing operations make forgery detection very difficult. In this work, a technique has been submitted to detect the effect of post-processing operation for forensic analysis. In this approach, the process has been break into three phases. In the first phase, preprocessing operation is performed where RGB image is converted into YCbCr image and extract the Cb and Cr image component for further processing. The second phase is feature extraction, multi-level DWT is implemented over chrominance component of the image and Canny edge detection technique is used to detect the edge of the image to localized the forgery, Otsu s based enhanced local ternary pattern (OELTP) technique is implemented on image to detect forgery-related artifact or feature of image. Edge texture is used to improve the performance of the Otsu global threshold. The third phase is classification, where Support vector machine (SVM) is used to classify the features of the image to find the image is forged or not. The accuracy of the proposed work is 99.98% on CASIA v1.0, 99.03% on CASIA v2.0, and 98.56% on the Columbia image dataset. The Sensitivity Rate also known as True Positive Rate (TPR) of the proposed model is 99.23% in CASIA v1.0, 99.06 in CASIA v2.0, and newline98.99% in COLUMBIA image dataset. The Specificity Rate also known as True Negative Rate (TNR) of the proposed model is 99.02% in CASIA v1.0, 98.68 in CASIA v2.0, and 9.23% in COLUMBIA dataset. The accuracy of the existing work is 98.65% on CASIA v1.0, newline98.01% on CASIA v2.0, and 97.25% on the Columbia image dataset. newline