Applying neural networks in coordinated group signal transformation to improve image quality

• Published in: Kompʹûternaâ optika Vol. 48; no. 6; pp. 924 - 931
• Main Authors: Lopukhova, E.A., Voronkov, G.S., Kuznetsov, I.V., Ivanov, V.V., Kutluyarov, R.V., Sultanov, A.Kh, Grakhova, E.P.
• Format: Journal Article
• Language: English
• Published: Samara National Research University 01.12.2024
• Subjects: energy efficiency; image processing; machine learning
• ISSN: 0134-2452; 2412-6179
• DOI: 10.18287/2412-6179-CO-1431

The rapid development of the Internet of Things and wireless sensor networks combined with the introduction of image analysis systems and computer vision technologies has led to the emergence of a new class of systems – multimedia Internet of Things and multimedia wireless sensing networks. The combination of the specifics of the Internet of Things, which requires simultaneous and long-term operation of a large number of autonomous devices, with the need to transmit video data poses the problem of creating new energy-efficient methods of image compression. The paper considers applying coordinated group signal transformation as such an algorithm, which performs compression based on the input signals’ correlation. Correlation between the image color channels makes this possible. However, it was necessary to supplement this method by clustering the original images using machine learning methods for better image reconstruction at reception. The criterion for clustering was the change of image gradient. The use of radial neural network in the clustering algorithm increased the speed of the proposed method. The resulting algorithm provides at least fourfold image compression with high-quality image restoration. Moreover, for multimedia Internet of Things systems, in which quality losses are acceptable, it is possible to provide large compression ratios without increasing computational complexity, i.e., without increasing power consumption.