The Study Of Remote Sensing Image Fusion Algorithm Based On The Wavelet Transform

As the development of the sensor technology, there are more and more remote sensing image data which are different in the time and space frequency. The multi-resource RS image fusion is becoming to be the very important research direction in the RS image processing areas. The wavelet theory is a new mathematical theory, especially when the famous mathematician Mallat found the high-speed algorithm of decomposition and reconstruction; it is used in many areas such as signal processing, image processing and so on. The algorithms based on wavelet theory and the combinations of the wavelet and the traditional theories are even hot in these years. This thesis is major in the fusion algorithm of the high space resolution panchromatic RS image and the low space resolution multi-spectral RS image.First of all, this thesis summarized the evaluation methods of the RS image fusion, it is included that the subjective evaluation and the objective evaluation. Secondly it summarized the traditional fusion algorithms such as the algorithms based on the Brovey transform, weighted method, HPF, IHS, PCA and the pyramid transform, and gave the evaluation of each algorithm. Thirdly, it sum up the algorithms based the wavelet transform, the wavelet packet transform, the multi-band wavelet transform and the combination of the wavelet transform and the traditional fusion methods such as the Brovey, IHS, PCA transform.Based on these fusion algorithms, a new RS image fusion algorithm is introduced which is based on the best tree transform in this thesis. Firstly, the panchromatic RS image and the multi-spectral RS image are decomposed using the wavelet packet transform. Secondly, every component is decomposed based on the entropy theory which is that, if the sum of the entropy of all son nodes is less than entropy of their father node, these son nodes are reconstructed. Thirdly, every component of the panchromatic RS image is merged with the corresponding component of the multi-spectral RS image. At last, the merged components were reconstructed using the inverse best tree transform. The fusion image based on this algorithm is wonderful according to the subjective evaluation, and it also considers the standard of objective evaluation. So it is certainly to say that this algorithm is a hopeful fusion algorithm. But the algorithm based on the wavelet transform are in the beginning stage, and the computational complexity is becoming more large form the wavelet transform to the wavelet packet transform and also to the best tree transform. To get more hopeful fusion algorithms which are superior in the evaluation and the computational complexity, more things should be done, such as construct new wavelet base which is more suitable for the image, and introduce the artificial neural network to the image fusion. These are the areas which should be further studied and explored.