Sparse Representation-based Multi-focus Image Fusion in a Hybrid of DWT and NSCT

摘要

This paper presents a sparse representation (SR)-based multi-focus image fusion by combining nonsubsampled contourlet transform (NSCT) and discrete wavelet transform (DWT). In this fusion framework, the Multi-scale transform (MST) is firstly applied on each of source images to acquire their low-pass and high-pass bands. Then, the low-pass coefficients are combined with a sparse representation-based fusion approach, while the high-pass coefficients are merged according to their absolute values as an activity level computation. Different MSTs with distinct decomposition levels can be utilized in this view of image fusion; however, NSCT-SR with one level of decomposition is best-performed multi-focus fusion. But it is more time-consuming compared to the other transforms. To overcome this defect, we combine DWT and NSCT in the first step. Since the size of low-pass approximate is decreased in DWT process, total fusion time is reduced. Moreover, in order to achieve better results, consistency verification is applied in low-pass fusion. By comparing the fused results subjectively and objectively, it is demonstrated that an acceptable trade-off of execution time and performance has been made.