Local Energy based Image Fusion in Sharp Frequency Localized Contourlet Transform

摘要

Image fusion method based on multiscale transform (MST) is a popular choice in recent research. Sharp frequency localized contourlet transform (SFLCT) that significantly outperform the original contourlet transform is proposed. Commonly, the upsamplers and the downsamplers presented in directional filter banks of SFLCT make the resulting image not shift-invariant and easily cause the pseudo-Gibbs phenomena. In order to suppress the pseudo-Gibbs phenomena, we apply cycle spinning as compensation. Then, the coefficients of shifted images are calculated. We take the following image fusion rules. First, cycle spinning the source images, the shifted images are obtained. Second, selecting the low-frequency coefficients by the local energy method and calculating the high-frequency coefficients by the sum modified Laplacian (SML), and the coefficients fusion follows. Third, applying the inverse SFLCT and the inverse cycle-spinning sequentially, the image is reconstructed. Numerical experiment results show that the proposed method significantly outperform the wavelet transform, the pyramid transform and the curvelet transform both in visual quality and in quantitative analysis.