Research on Removing Image Noise and Distortion in Machine Dial Recognition

Abstract

This study aims to address the issues of image noise and distortion in machine dial recognition via initial denoising. A wavelet local threshold denoising method that combines high-frequency wavelet coefficients with wavelet decomposition coefficients in various directions is proposed. This method shows good results on 102 images of car dashboards, aircraft instrument panels, spacecraft displays, and dial instruments on robots. Although a few denoised images exhibit distortion due to intense lighting or heavy contamination, the denoising accuracy for the remaining images is 98.04%, demonstrating substantial practical value. Future research will concentrate on addressing complex image noise and structures.