Performance Comparison of Image Fusion Alternatives Combining PCA with Multi-resolution Wavelet Transforms

In the situations of image fusion between the panchromatic (PAN) and the multispectral (MS) image, it is an ideal thing to properly absorb the source image ingredients to reach the best state of the space clarity and the spectrum feature in the fusion image. Principal component analysis (PCA), Contourlet and Curvelet methods are easy to bring about serious color loss during the process of image fusion. The method using PCA and multi-resolution wavelet transform (MWT) with the corresponding fusion rules can be used to remedy the faultiness. MWT includes non-decimated wavelet transform (NDWT), wavelet package transform (WPT) and lifting wavelet transform (LWT) in the study. The main ideas of this study embody the three aspects. Firstly, the original MS image is converted by using PCA to get the first principal component: PC1. Secondly, NDWT, WPT and LWT are applied to decompose the PAN image and the PC1 image, respectively. A new PC1 can be reconstructed after the reversed wavelet transform when the different fusion rules are adopted in the low- and the high-frequency domains separately. Lastly, the inverse PCA transform is performed by combining the new PC1 and the other two components of the MS image. In terms of the same series of MWT, the experimental criteria perform as follows. The proposed fusion rule of each MWT gets the best value of the SSIM, which are 0.7747, 0.7696 and 0.7504 for the first group of data and 0.7747, 0.7696 and 0.7504 for the second group of data. The OP criterion performance ranks the second; and other criteria highlight the advantages of the fusion rule. By comparing with other fusion methods, there are three points concluded as follows. Firstly, fusion methods based on wavelet analysis are better than those based on super-wavelet analysis in some image fusion situations. Secondly, mutual comparison among these three fusion frames which refer to PCA combining with NDWT, WPT and LWT, respectively, can lead to the fusion performances including spatial clarity and spectral loss under various rules. Thirdly, the proposed fusion rule can play the maximum complementarity of the two aspects when combining PCA with WPT.