Image tampering blind forensics method based on detection-segmentation architecture

Abstract

The invention discloses an image tampering blind forensics method based on a detection-segmentation framework, and belongs to the field of deep learning and computer vision. According to the method, the tampering detection problem of an image is solved from the perspective of semantic segmentation, and tampering region detection and region segmentation are carried out by adopting an improved Mask R-CNN network. In a data preprocessing stage, original pictures are subjected to data enhancement, and sample richness is enhanced. In a feature extraction stage, a path from bottom to top is newly added to achieve fusion of multi-level feature information, and sufficient context semantic information is further obtained. In an RPN training stage, Focal Loss is adopted to solve the problem of imbalance between positive and negative samples, so that the network is easier to converge. In a post-processing stage, a Soft-NMS algorithm is adopted to solve the problems of false detection and missing detection caused by overlapping detection frames, and the detection recall rate is improved. According to the method, blind forensics of various tampered (mode) images in a complex scene is realized, and the detection and positioning precision of a tampered area is effectively improved.

Description

Technical field [0001] The present invention relates to the field of digital image forensics, computer vision, and digital image processing, and is specifically a method of authenticity and integrity detection of digital images without any prior information. Background technique [0002] Digital image blind tailor technology is directly delivered to the image content in the field of digital image forensics. With the development of computer technology and information technology, the image is getting more and more popular, and the tampering content is getting more and more, the tampering means is increasing, causing the difficulty of challenge the image to the image. Therefore, it is necessary to study the strong blind test algorithm, which can accurately perform authenticity and integrity detection of tampering images. [0003] Traditional image tampering blind drawback method is mainly a truly detection of image of the "handmade design feature + classifier". Such methods solve if...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to propose a blind image tampering forensics algorithm based on detection-segmentation architecture for the problems of inaccurate tampering area positioning and solidified detection methods in the existing deep learning-based image tampering blind forensics algorithm

Images