Abstract: A wavelet-packet fusion algorithm based on regional edge features is proposed, to solve the disadvantage of traditional wavelet transform that it hasn’t decomposed the high-frequency components further and ignored the details. In this method, energy-weighted fusion rule is applied for the low-frequency components after wavelet packet decomposition, while high frequency components calculates the edge feature statistics, utilizing the directionality of each sub-band, and gets fusion coefficients by weighted method. The experiments on SPOT multi-spectral image and high-resolution image are carried out. The experiment data and theory analysis show that the average gradient evaluation parameters etc of this method are raised, and it effectively improves the spatial feature information of the image while maintaining the spectral information.

Key words: wavelet packet transform, edge feature statistics, sub-band direction, image fusion

摘要： 针对传统小波变换融合方法中未对高频分量作进一步分解从而忽略了细节信息的缺点，提出了一种基于区域边缘特征的小波包融合算法。该方法对小波包分解后的低频分量采用能量加权的融合规则，高频分量则利用各个子带的方向性，计算其边缘特征统计量，通过权值法得到融合系数。对SPOT多光谱图像和高分辨率图像进行融合实验，实验数据和理论分析表明，该方法的平均梯度等评价参数均有提高，在保持光谱信息的同时，有效地改进了图像的空间特征信息。

关键词: 小波包变换, 边缘特征统计量, 子带方向, 图像融合