A new approach based on likelihood function is proposed to detect ships. A window structure of the neighborhood is constructed by using the idea of non-local filtering. For non-center widows, the repeating neighborhood windows are employed to estimate covariance matrices corresponding to the different positions of the neighborhood window. And then likelihood functions of the covariance matrices and the neighborhood window of center pixel are employed to compose the ship detector. The effectiveness of the proposed algorithm is proved by using the Polarimetric Synthetic Aperture Radar(PolSAR) data and is compared with existing algorithms, which demonstrates that the proposed algorithm can obtain better experimental results.