Image analysis was used to evaluate the curve grinding process. The measurement system organization and its principle of operation were introduced. Some key technologies that influence the system precision were also studied in details. Real-time image of work piece and wheel grinding can be gathered while using CCD camera. For image de-noising, a kind of wavelet threshold function was presented to calculate the new wavelet coefficients. Local threshold algorithm was used to compute different scale threshold, another threshold is suggested when MSE is minimum. Image signals are reconstructed through invert discrete wavelet transformation with the new wavelet coefficients. For the purpose of increasing the measurement precision, a closed mathematical model of sub-pixel edge detection operator was founded according to the gray moment operator’s invariant characteristics and Tabatabai’s first three sample gray moments. An application of system validity and the comparison results of work piece were also given. From experiments it can be said that the proposed method in this paper is effective, and its detection precision and results are reasonable.