A security JPEG image system accelerated by NEON technology based on FT-2000/4

• Published in: CCF transactions on high performance computing (Online) 2024-10, Vol.6 (5), p.488-502
• Main Authors: Hu, Yu, Li, Ziteng, Li, Jianfeng, Tie, Junbo, Wang, Lei
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Compression ratio; Computer Hardware; Computer Science; Computer Systems Organization and Communication Networks; Confidentiality; Embedded systems; Explosive compacting; Hardware; Image compression; Image manipulation; Image transmission; Libraries; Mathematical functions; Neon; Neural networks; Quality standards; Real time; Regular Paper; Security
• ISSN: 2524-4922; 2524-4930
• DOI: 10.1007/s42514-023-00177-5

With the development of network bandwidth and the explosive growth of information, the number of users transmitting images to the public cloud for storage is surging, stimulating the demand for real-time image transmission, processing, and secure storage. Processing a large amount of image data takes a lot of software and hardware resources; thus, image compression technology is essential. Simultaneously, protecting the privacy and security of the image during transmission and storage is the key to the introduced application scenario. However, it is hard to meet the real-time requirements by using Single Instruction and Single Data algorithms. To fix the issues, the focus of this work is on the implementation of JPEG compression with NEON technology on the FT-2000/4 platform, alongside the development of a system named IMAGE-DETECTIVE. The goal is to address the challenges associated with real-time image transmission, processing, and secure storage. The open-source library libjpeg-turbo is utilized to achieve efficient compression and decompression. To detect image tampering, a convolutional neural network is proposed, taking into consideration the “double compression ratio” characteristic often observed in tampered JPEG figures. Additionally, the FT-2000/4 platform’s hardware SCTO module is employed for JPEG image encryption. In addition, we employ the NEONMATH, a NEON technology math library we proposed, to accelerate the process of JPEG image pre-processing and training phase of the neural network. Our system improves the speed of JPEG compression and decompression by 8.72×, and the accuracy of tampering detection reaches 93.3%. The code is available at https://github.com/nudtyuyu/IMAGE-DETECTIVE .