| abstract |
The invention discloses a remote sensing image fusion method based on weighted average curvature filter decomposition, which adopts an adaptive weighted average method to fuse each band of a multispectral image to generate an intensity component I. Spectral estimation is performed with the intensity component I as the initial alpha channel, and the foreground color F and background color B are obtained. The panchromatic images are decomposed into small-scale images, large-scale images and base images by weighted average curvature filtering and Gaussian filtering. The size of the connection coefficient is dynamically adjusted by using the local energy of the image. The intensity component I and the base image are fused by a pulse-coupled neural network. The fused images, large-scale images, and small-scale images are linearly combined to construct the final alpha channel. According to the image matting model, the foreground color F, background color B, and the final alpha channel are reconstructed to obtain a fusion image. By comparing with ten typical fusion methods, the present invention achieves better results in both subjective visual effect and objective evaluation. |