The phase measuring deflectometry fails to measure the double surfaces of transparent objects, because the fringes reflected from both the front and rear surfaces are superimposed in the captured images. Therefore, accurately extracting the phase from the coupled fringe patterns is considered as one of the main problems in PMD. A fringe decoupling method based on continuous wavelet transform is proposed to eliminate parasitic reflections. It proposes a method based on modular maximum and image segmentation, which extracts the best multi-region ridge points in time-frequency domain and obtain the principal components of different wavelet ridges in amplitude spectrum. Then the fringe phase of the corresponding position of the wavelet ridge is extracted. The proposed method successfully attains a curvature radius mean value of 0.038mm for 135.800~515.090mm radius of curvature lens in the experiment. The root mean square of the surface shape error is 1.4μm. The proposed algorithm can eliminate parasitic reflections successfully which eventually realizes high-precision measurement of transparent objects.