Abstract: In order to improve the effect of gear fault detection and defect detection, and effectively remove the mixed noise in the image, an edge detection algorithm of noisy gear image combined with image enhancement was proposed. First, the information entropy was introduced to improve the Mahala Nobis distance formula, and the Mahala Nobis distance was used to improve the adaptive median filter. Then the power transform was improved to make it adaptive, and the improved power transform was used to improve the Retinex algorithm to enhance the overall image. Finally, the wavelet modulus maximum method was used to detect the edge of the noisy gear image. The research results show that the objective evaluation indexes PSNR, SNR and SSIM of this algorithm are higher than the objective evaluation indexes of median filter and adaptive median filter for mixed noise, and the denoising effect is significantly improved. At the same time, combining with the improved Retinex algorithm, the enhanced brightness and contrast of the overall noisy gear image are increased, and part of the noise is suppressed, making the edge detection effect of noisy gear image better.

Key words:  mixed noise; Mahala Nobis distance; Retinex algorithm; wavelet modulus maximum