To extract more detail information of the image and enhance the calculation effectiveness, an improved non-subsample shearlet transform(NSST) is proposed. The improved non-subsampled pyramid(NSP) in the traditional NSST is replaced by the redundant lifting non-separable wavelet. In this improved NSST, the low frequency subbands are fused by summing of the region energy, while the high frequency sub-bands are optimized by a simpled-pulse coupled neural network(PCNN) model. The experimental results show that the proposed method is of very competitive advantages in extracting exact information from images according to evaluation criterion of both visual appearance and objective criteria.