A team from the University of Macau (UM), led by Prof Zhou Jiantao in the Faculty of Science and Technology, has successfully developed a new algorithm for digital image tampering detection. With this technology, the team competed with 1,534 teams from around the world at the Security AI Challenger Contest (Season 5) co-organised by Tsinghua University and Alibaba. The team won the championship in the tampering track and a third prize in the detection track, making UM the only university in Macao to have received awards in this event. In addition, the team has been invited by Alibaba to participate in one of its innovative research projects, which will help to promote the commercialisation of scientific research results and academia-industry collaboration.
In recent years, with the popularisation of photo editing applications, a large number of forged images have been used to spread rumours, fabricate news, or illegally obtain economic benefits. Against this backdrop, image tampering detection technology has been widely adopted to detect forged images in fields such as media, justice, and criminal investigation. In order to improve the existing detection technology and promote the application of image forensics technology, Tsinghua University and Alibaba jointly held the world’s first tampering detection contest for image forgery on certificates and ID cards.
During the competition, UM conducted a detailed analysis of image tampering and the detection process, and proposed a new tampering algorithm by alternating true and false areas and adding adaptive noise to generate forged images that are difficult to detect through the senses. In addition, UM proposed a new detection algorithm based on deep learning technology. The algorithm adopts a multi-network architecture that integrates spatial channel perception modules to recognise the tampered area of a forged image. It uses the GPU supercomputer provided by UM’s Super Intelligent Computing Center (SICC) to detect the input image in seconds. The team’s detection algorithm can locate the forged area of a tampered image within 0.5 seconds.
The supervisor of the UM team is Prof Zhou Jiantao, an associate professor in the State Key Laboratory of Internet of Things for Smart City (SKL-IOTSC) and interim director of the Center for Artificial Intelligence and Robotics. The team members are Wu Haiwei and Sun Weiwei, two doctoral students in the SKL-IOTSC, and Xiong Ziyi, a master’s student in the Faculty of Social Sciences. Because of the team’s outstanding results, Prof Zhou and Wu have been invited by Alibaba to participate in an innovative research project, titled ‘Research on Robust Anti-Media Transmission Forgery Image Detection and Localization. They have also received a sponsorship of RMB 500,000 from Alibaba. This project aims to promote academia-industry collaboration and serve as a successful example of collaboration between universities and IT companies. It is also expected to help UM develop a practical and efficient AI algorithm, enhance its skills in real-world environments, and nurture outstanding professionals with practical experience.
*Prof Zhou Jiantao was interviewed by local media TDM, please click below link to see the full version of the video
https://www.tdm.com.mo/c_news/tv_news.php?id=517482
澳門大學科技學院教授周建濤帶領的團隊成功研發出一項新的圖像篡改檢測技術算法，並以此參加由清華大學和阿里巴巴主辦的全球首個針對偽造證件類圖像的檢測比賽“安全AI挑戰者賽（第五期）”，與全球1,534個團隊相互切磋，最終獲得篡改賽道冠軍及檢測賽道季軍佳績，是全澳唯一獲獎的高校。此外，團隊更獲阿里巴巴邀請和贊助參與其創新研究計劃專案，促進科研成果轉化，共同推進“產學研”發展。
隨著各種修圖工具的普及，近年大量篡改圖片被濫用作散播謠言、編造新聞、非法獲取經濟利益等。而圖像篡改檢測技術正好用於檢測篡改的圖像，目前已在新聞、司法、刑偵等領域廣泛應用。但隨著反取證技術的出現，圖像的真偽鑒定問題進入了攻防雙方博弈的階段。為了提升現有檢測技術及推進圖像取證技術的應用，清華大學和阿里巴巴合辦全球首個針對偽造證件類圖像的檢測比賽。
比賽期間，澳大對圖像的篡改以及檢測流程進行了詳盡的分析，包括提出了新的篡改算法——通過交替真假區域以及添加自適應性噪聲來生成難以通過感官辨識到的偽造圖像；同時，也提出了新的檢測算法——基於深度學習技術，使用融合了空間通道感知模組的多網路架構對偽造圖像進行篡改區域的定位檢測。此算法運用了澳大智能超算中心（SICC）所提供的GPU超級計算機，以秒級單位對輸入圖像進行檢測。澳大團隊的檢測演算法可於0.5秒內對被篡改圖像進行偽造區域定位。
周建濤亦是澳大智慧城市物聯網國家重點實驗室副教授、人工智能與機器人研究中心代主任。隊員為澳大智慧城市物聯網國家重點實驗室的博士生吳海威和孫巍巍，以及澳大社會科學學院的碩士生熊子儀。憑藉出眾的實力，周建濤和吳海威獲阿里巴巴邀請參與其創新研究計劃專案“抗媒體傳輸的高魯棒偽造圖像檢測與定位研究”，並獲贊助 50萬元人民幣。此專案成功促進“產學研”的融合，為院校及互聯網企業的合作提供更多經驗及參考；有助澳大建立實用高效的AI演算法，在高難度的真實環境中提升技術，培養有實戰經驗的優秀人才。
*周建濤教授接受本地媒體TDM採訪，請點擊以下鏈接查看完整版視頻
https://www.tdm.com.mo/c_news/tv_news.php?id=517482
