Abstract: In order to improve the effect of image fusion, an image fusion algorithm using spiral structure and gradient analysis is proposed. First, a series of high and low frequency subgraphs are obtained by nonsubsampled contourlet transform. Then, the sliding window model for the sparse representation method in the low-frequency subgraph is studied. In view of the slow fusion time, the spiral structure direction model is proposed for dictionary learning and sparse representation. In this model, the low-frequency subgraph is fused based on the rule that the sparse coefficient is determined by the largest spatial frequency, which improves the fusion efficiency. Aiming at the problem of edge mutation of the fused image in the high frequency subgraph, the high frequency fusion rule based on gradient analysis is presented to make the clearer image features easier to be retained in the final fusion image. Finally, the fusion experiments for gray and color images are carried out and the comparison and analysis between different fusion algorithms are demonstrated. The validity of the proposed algorithm, including time and fusion effects, is verified through the comparison results of subjective observation and objective data.

Key words: image fusion, sparse representation, spiral structure, gradient analysis

摘要： 为了提高图像融合的效果，提出了螺旋结构和梯度分析的图像融合算法。算法首先进行非下采样轮廓波变换，得到一系列高低频子图。然后对低频子图中稀疏表示方法的滑窗模型进行了研究，针对其融合时间较慢的问题，提出了螺旋结构方向模型进行字典学习和稀疏表示，对稀疏系数通过空间频率取大的规则进行低频子图的融合，提高了融合效率；又针对高频子图中待融合图像的边缘突变情况，提出基于梯度分析的高频融合规则，使得较清晰的图像特征在融合时更易保留至最终的融合图像中。最后，对灰度图像和彩色图像进行了融合实验及不同融合算法的比较分析，并通过主观观察和客观数据对比验证了该算法在时间上和融合效果上的有效性。

关键词: 图像融合, 稀疏表示, 螺旋结构, 梯度分析