An Image Fusion method Based on PCA Cosparse Representation

Abstract Pan-sharpen as a tool of remote sensing image resolution enhancement has been intensively researched by scientists in the past few years. The sparse based methods have obtained desirable results. Unfortunately, many approaches require large numbers of high resolution images that acquired from the same sensors to train the dictionary. That often spends the computational burdens largely. For other sparse based methods, it is necessary for the spectral response of the PAN (Panchromatic) image to comprehensively encompass all bands of the MS (Multispectral) image with constructing the coupled dictionary, which is usually useless for some optical sensors such as WorldView-2. In this work, a new Pansharpening method was proposed based on principal component analysis (PCA) cosparse representation. The analysis operator was constructed with PCA subspace avoiding analysis operator learning procedure that reduces computational resources. Moreover, in this work, the possible signal structure correlations between multispectral channels are fully exploited, which overcomes the spectral discrepancy between the Pan and multispectral images. The algorithm is evaluated by simulated experiment with WorldView-2 data set. The result demonstrates that the proposed fusion algorithm is effective and superior to the other state-of-the-art image fusion approaches.