Abstract: Abstract: The article proposes high-order balanced multi-band multiwavelet packet transforms and the remote sensing image fusion based on the transforms for improving the performances of wavelet transforms based remote sensing image fusion. The properties of different wavelet transforms and the relationship among them were firstly analyzed. It was shown that although the balanced multi-band multiwavelet transform had improvements in signal processing due to the properties of balancing and multi-band systems, it should be developed to its wavelet packet counterpart for more effective spatial-frequency domain representation and further improvement in application. Then, the principle theory of the high-order multi-band multiwavelet packet transform was analyzed and it was found the library of the base functions of the transform is an M-fold tree structure (M is the band number of a wavelet system). The fast algorithms of the decomposition and reconstruction of the transform were presented. With the algorithm, one could obtain the transform-domain coefficients in different frequency bands of an analyzing dataset. Correspondingly, the library constituted by these coefficients could be also viewed as a set with M-fold tree structures. Using a tree-information searching algorithm based on cost functions, one could find the most suitable sub-tree library in such a tree-structure set. Based on the coefficients corresponding to the library, the high-order balanced multiwavelet packet transform, which was more suitable to the dataset, could be established. Next, the remote sensing image fusion method based on the transform scheme was constructed. An existing simple fusion strategy was used to construct the wavelet transform based fusion method so as to present a clear evaluation on the function of wavelet transforms. The fusion strategy integrated the intensity-hue-saturation (IHS) transform and wavelet transform together. The IHS transform was used to obtain the intensity component of the multi-spectral images in a remote sensing image dataset. After that, the obtained intensity component and the panchromatic band in the dataset were both decomposed at different levels by the high-order multi-band multiwavelet packet transform, and the coefficients corresponding to different frequency bands were then generated. After the most suitable coefficient sub-tree library for the dataset was obtained, the low and relatively high frequency coefficient parts in the library were then merged with average fusion rule and maximum fusion rule respectively. After performing the reconstruction of the proposed transform on the merged coefficient, the merged intensity component was obtained. Finally, the fusion image could be generated after an inverse IHS transform. With this fusion strategy, this paper comparatively evaluated the performance of the high-order balanced multi-band multiwavelet packet transform based fusion method and other traditional wavelet packet transform based fusion methods. In the experiment, these methods were applied to the image fusion between the panchromatic bands and multi-spectral bands of the datasets received from different satellites, such as Chinese Resource Satellite and Landsat-7 Satellite. These datasets containing rich textures and spectral information, could present an effective analysis. The experimental result showed that for different band number M, the proposed method generally gains the best fusion result when balance order ρ equals to 2, and in different balance order ρ, the proposed method with band number 3 obtains the best result in most experimental dataset cases. Also, it showed that in general, the suitable decomposition level for the proposed method is 2 or 3. Moreover, it was found that with a suitable balanced order and band number, the proposed method could provide improvements in both visual quality and objective evaluation compared with traditional fusion methods. Especially, it could reduce both the average spectral error and relative global dimensional synthesis error by more than 3 percent compared with traditional methods. Also it incurred reasonable computational complexity compared with traditional methods. The method is thus useful for remote-sensing image fusion. Furthermore, it is also advantageous for many other image processing tasks, such as texture extraction and edge detection.