A super Gaussian model is established to describe the characteristics of real fiber Bragg grating (FBG) reflective spectrum. Utilizing image processing technique, a FBG peak detection technique based on Steger image algorithm is proposed, which provides a modified formula for the condition of asymmetric spectrum and a sub-step modified formula. Steger peak detection algorithm performs the function of both filter and peak detector to help eliminate the influence of noise, step and asymmetrical spectrum. Theoretical simulations and practical experiments are carried out to estimate the performance of this algorithm under different steps, signal noise ratio (SNR) and level of asymmetry. The results show that the proposed algorithm shows better adaptability and higher accuracy than the Centroid method and Gaussian curve fitting.