Consistent pan-sharpening based on multistage joint and dual bilateral filters

A pan-sharpening method using joint and dual bilateral filters (DBFs) has been proposed. This approach is based on a consistent combination of large- and small-scale features obtained from the decomposition of high spectral resolution multispectral (MS) and high spatial resolution panchromatic (PAN) images. In the decomposition process, MS and PAN images are used to extract the features using joint and DBFs, respectively. These filters accommodate the relationship between MS and PAN images and decompose them into a base layer (large-scale) and a detail layer (small-scale). Since the joint bilateral filter (JBF) preserves the edges of an auxiliary image, it is used for decomposition of MS images where different layers are estimated using the PAN image as an auxiliary image. Similarly, different layers of the PAN image are obtained from a DBF which preserves the edges of both (MS and PAN) input images. This process is further extended to multistage decomposition to obtain a bilateral image pyramid. The base and detail layers of both MS and PAN images obtained at various stages are combined using a weighted sum. Finally, the estimated weighted sum of detail layer (small-scale) of the PAN image is fused separately to the weighted base layers (large-scale) of the MS images. Performance of the proposed method is evaluated by conducting the experiments on degraded as well as undegraded datasets, captured using different satellites such as Quickbird, Ikonos-2, and Worldview-2. The noise rejection capabilities of the proposed method are also tested by conducting experiments on the noisy data. The results are compared with the widely popular methods and the recently proposed fusion approaches based on a bilateral filter. Along with qualitative evaluation, the quantitative performance of the proposed fusion technique has also been verified by estimating different measures for degraded and undegraded experiments. The experimental results and quantitative measures demonstrate that the proposed method performs better in degraded and undegraded conditions along with noisy situations when compared to other state-of-art methods.