基于非下采样Contourlet变换的多分辨率图像融合方法

Multi-resolution Image Fusion Based on the Nonsubsampled Contourlet Transform

摘要:提出了一种基于非下采样contourlet变换(NSCT)的多分辨率图像融合方法,通过非下采样金字塔(NSP)和非下采样方向滤波器组(NSDFB)实现对图像的多尺度多方向分解.该方法既保留contourlet变换方法良好的多分辨率特性,又具有平移不变性.在融合处理中,采用一种改进的一致性校验策略;在高频系数中除了进行本层的一致性校验外,还进行多层之间的一致性校验.实验结果表明,该方法取得了良好的融合效果,所得融合图像的多项指标都优于拉普拉斯金字塔变换、小波变换、contourlet变换等方法.

An image fusion method based on nonsubsampled contourlet transform （NSCT） is presented, which uses the nonsubsampled pyramids （NSP） and the nonsubsampled directional filter banks （NSDFBs） to decompose the source images in a multi-scale and multi-direction way. Like the contourlet transform method, the presented method can obtain multi-scale and multi-direction images with good multi-resolutions, and it is also shift-invariant. In the fusion process, an improved consistency check strategy is used. The high frequency coefficients are checked not only in their own level but also in all high frequency levels. The experimental results show that the presented fusion method is effective, and it can obtain better fused images than such methods as Laplacian pyramid transform, wavelet transform and contourlet transform with respect to several indicators.