S-transform profilometry is a three-dimensional shape reconstruction method based on a lossless and reversible time-frequency technology. This method, a multiresolution technique, can reconstruct the three-dimensional shape of the tested object using the phase information demodulated from a single-shot fringe pattern. Herein, we analyze the factors that may affect the accuracy of S-transform profilometry. Piecewise-mean and curve-fitting methods are proposed to eliminate the background intensity of the fringe. In addition, adjusting factors are introduced into the S-transform kernel function to improve the time-frequency resolution. Simulation and experimental results verify that the proposed method exhibits high accuracy of three-dimensional shape reconstruction because of accurate S-transform coefficients.