Image fusion is the combination of two or more images to obtain a new image of better quality using a certain algorithm. Due to physical or observational constraints, single sensor or single image can't afford enough information. By the synthesization of multisensor image, the obtained information is more reliable and robust and the range and the ability of information is expanded. So image fusion benefits image understanding and scene interpretation. Image fusion is now extensively applied to many areas, such as remote sensing, medical imaging, digital camera, defense and military, and so on.Multiscale image fusion algorithm plays an important role in the image fusion algorithm at pixel level. However, these methods don't recognize the difference of image scales at the scale space. The difference will lead to more artificial information in the fused result. Considering the difference in image scales, an image fusion idea is first proposed that image fusion should performed at similar scale. The idea improves the combination of low resolution multispectral image and high resolution panchromatic image.According to the idea of similar scale, an image fusion algorithm based weighted multiscale fundamental form at similar scale is proposed. Firstly, we use the "à trous" discrete wavelet transform to decompose the panchromatic image so that its approximation and the interpolated multispectral image have the similar scale. Secondly, image fusion based on weighted multiscale fundamental form is performed on them to obtain the synthesized approximation. At last, the inverse "à trous" wavelet transform of the synthesized approximation and the detail of the panchromatic image reconstructs a high spatial resolution multispectral image.Another image fusion algorithm at similar scale is proposed using a variational approach. The algorithm performs image fusion through energy minimization. The energy functional mainly consists of two components. One ensures the injection of more correlated detail spatial information. The other ensures the preservation of the spectral information. It is used to preserve the spectral information that a lowpass approximation of the enhanced multispectral image at a determined level and the multispectral image to be fused should have the similar scale. Additionally, prior knowledge about the high spatial resolution multispectral image and other constraints can also add to the energy.Additionally, this thesis also makes the following contributions: weighted multiscale fundamental form is proposed and is applied to image fusion, image enhancement, anisotropic diffusion and color image restoration; the performances of empirical mode decomposition and wavelet decomposition in the field of image fusion is compared; the combination of spotted IC image and the object image using the fuzzy integral is used to restore the spotted region.
修改评论