Abstract: While a stripe laser vision sensor is used to measure the welded forming size, the noise is often existing in a weld image. Aiming at the deficiencies of hard threshold function and soft threshold function in traditional wavelet threshold denoising algorithm, a denoising algorithm based on Soft-hard Threshold Compromise method is studied. Using db4 wavelet basis function, three layers with wavelet decomposition of laser stripe center are performed. The traditional wavelet threshold function and Soft-hard Threshold Compromise method are applied to denoise the detail coefficients, and the denoising effects of several methods are evaluated by SNR and RMSE. Experimental results show that the Soft-hard Threshold Compromise method is better than the traditional wavelet threshold method for denoising, which can effectively improve the measurement accuracy and stability of weld forming size.