PR‐NET: Progressively‐refined neural network for image manipulation localization

Abstract

Current deep learning‐based image manipulation localization methods achieve impressive performance when rich spatial features and information are fully utilized. However, most of them suffer from the irrelevance of semantic awareness when identifying various manipulation categories. This leads to false alarms on recognizing forged regions. In this paper, we propose a Progressively‐Refined Neural Network (PR‐Net), to localize the tampered regions progressively under a coarse‐to‐fine workflow. Specifically, PR‐Net is composed of a Feature Extractor (FE) that captures feature intrinsic correlations and a Mask Generation Module (MGM) with three refining generators. The FE takes a CNN to extract the image features and introduces an attention mechanism Convolution Block Attention Module (CBAM) to suppress the image content and guide the extractor in exploring the inconsistencies between the manipulated and authentic regions. The MGM comprises three generators where the Coarse Mask RR‐Generator generates a localization result roughly, the Candidate Mask RR‐Generator generates a possible tampered region according to the rough localization measure, and the Fine Mask RR‐Generator produces the final prediction of manipulated regions. We also utilize the Rotated Residual (RR) structure to suppress the image content during the generative process. The extensive experimental results on four benchmark data sets (NIST16, COVER, CASIA v1.0, and In‐The‐Wild) demonstrate the superior performance of PR‐Net compared with the state‐of‐the‐art methods in localizing the manipulated regions.

Progressive selection‐channel networks for image steganalysis

“…However, the semantic meaning of the modification probability map is quite weak, and it is difficult to generate precise maps directly using simple backbone networks. To this end, we design a multistage progressive network to estimate the probability maps in an "easy-to-hard" learning paradigm, whose recursive computation process [37][38][39][40] also significantly reduces the network parameters for fast inference.…”

A comprehensive study of Mozi botnet

“…With the large-scale construction and application of new-generation digital infrastructures such as 5G, industrial Internet, big data centers, and cloud computing, more and more important information systems will carry core businesses and massive amounts of data that are closely related to national security and economic development. 1,2 More and more researchers have begun to focus on artificial intelligence (AI) technology applications, [3][4][5][6] network infrastructure optimization, 7,8 and software security analysis 9,10 for critical infrastructure. The development and research results of these works also bring some inspiration for the study of the security of critical infrastructure for us.…”

“…(3) What is the unique identifier generation strategy for Mozi in the distributed hash table (DHT) network? (4) What is the difference between the communication methods of Mozi and infected nodes and normal nodes?…”

Single image deraining using multi‐stage and multi‐scale joint channel coordinate attention fusion network

“…To pay more attention to the relevant areas of the image to improve the performance on removing rain streaks, some attention modules are introduced. [10][11][12][13][14] Though these methods achieved state-of-the-art results, dividing the rain image into a rain part and a background part remains challenging. In the real world, there are rain streaks with different directions and densities, and these methods can only extract information about rain streaks in a single direction, and some rain streaks are very similar to the background, which leads to the loss of important background information while removing rain streaks.…”

“…There are also networks that use FPN 9 to remove rain on multi‐scale features. To pay more attention to the relevant areas of the image to improve the performance on removing rain streaks, some attention modules are introduced 10–14 …”