Abstract: in this paper, a new algorithm is presented to decrease the cross-terms in the Wigner distribution resulted from the bilinear character The separation between auto-term and cross-term is performed by using Hough transform, forming a dynamic kernel varying as signal changes. By using such a kernel, energy of the detected signal can be retained at quite a high level, thus the final time-frequency distribution can reflect the energy distribution of original signals, which is irregular in traditional Choi-Williams distribution or its improved form. Experimental results prove the effectiveness of new algorithm.