Analysis of image processing methods in the Internet of Things systems based on wavelet transformations

Abstract

Increasing the compression ratio of images to reduce their transmission time in sensor networks based on microcontrollers helps to increase the overall energy efficiency of the system. The purpose of the study was to investigate the effectiveness of using Haar, Daubechies, and Coiflet wavelet transformations for image compression on 32-bit microcontrollers. An experimental comparison of the efficiency of three types of wavelet transformations for processing images obtained from the built-in camera was performed by the metrics of root-mean-square error, peak signal-to-noise ratio, structural similarity index, and Euclidean distance. Haar, Daubechies, and Coiflet wavelet transform algorithms were implemented on the ESP32 microcontroller. The results showed that at the second level of decomposition, the Haar wavelet provided high image quality (MSE 25.153, PSNR 34.124 DB), but at the fourth level, the quality significantly deteriorated (MSE 73.449, PSNR 29.470 DB). The Daubechies wavelet showed similar results, but at the fourth level, its efficiency also decreased (MSE 78.241, PSNR 28.974 dB). Coiflet wavelet showed the worst results at the fourth level (MSE 89.630), but at the second level its quality was competitive. For the first time, three types of wavelet transformations were compared using an additional metric – Euclidean distance, which made it possible to better estimate artifacts and image distortions. The proposed approach allowed improving the efficiency of image compression and transmission in the Internet of Things systems on microcontrollers, which provides less data transfer time and, accordingly, reduces power consumption, which is critical for autonomous sensor networks.