Detection of Duplicate Region and Hybrid Non-Local Means Filtering for Denoising with Quantization Matrix Estimation for JPEG Error Analysis

Abstract

Background: The existing JPEG error analysis schemes do not offer agreeable results particularly when the duplicated area is small. Region duplication is an uncomplicated and efficient process to produce digital image forgeries, where a constant segment of pixels in an image, following feasible geometrical and illumination transformations are copied and pasted to a different location in the same image. Methods: In this research work, JPEG error analysis scheme is introduced for the purpose of consistent recognition of duplicated and distorted areas in a JPEG digital image forensics. Here, presented a new Multi-directional Curvelet Transform with Fourier Transform matching Invariant Rotation (MCFTIR) region duplication detection scheme to identify duplicated regions for JPEG images. This scheme begins with estimating the overlapping blocks of a JPEG image and it is organized in accordance with the statistics of multiple curvelet sub-bands. During the second phase, the amount of candidate block pairs of JPEG images has been significantly diminished by means of spatial distance between each pair of blocks for JPEG image. For duplicate region removed images theoretically analyzing the effects of these errors on single and double JPEG compression, with five major phases like Shape-Preserving Image Resizing (SPIR) scheme for the purpose of image resizing, noises are appended to image and eliminated with the help of Hybrid Non-Local Means Filtering (HNLMF) denoising framework, Image compression through Discrete Cosine Transform – Singular Value Decomposition (DCT-SVD) was computed for single and double image compression, images were quantized by means of numerous quantization matrices, quantization matrix results are estimated with Mamdani model based Adaptive Neural Fuzzy Inference System (MANFIS) and detecting the quantization table of a JPEG image. Findings: The proposed MCFTIR methods significantly outperform existing techniques in terms of the parameters like Peak Signal to Noise Ratio (PSNR) and Mean Square Error (MSE) especially for the images of small sizes. It also show that the new MCFTIR method can consistently identify JPEG image blocks which are as tiny as 8x8 pixels and compressed through quality factors as elevated as 98. This performance is significant for the purpose of analyzing and locating small tampered regions inside a composite image.