动态WNMF及在图像融合中的应用研究

Dynamic Weighted Non-Negative Matrix Factorization and Its Using Research in Image Fusion

标准非负矩阵分解图像融合算法全局特征提取能力有限,造成融合图像的对比度不高,视觉效果不好,针对这一问题,对加权非负矩阵分解算法进行了深入研究,提出了动态加权非负矩阵分解思想并将之应用于红外与可见光图像融合。动态加权非负矩阵分解算法首先通过加权系数的设计指定重要特征,并在迭代过程中根据各区域相对重要程度的变化对加权系数进行动态调整,与标准非负矩阵分解算法相比较,动态加权非负矩阵分解算法全局特征提取能力得到了显著提升。对比实验表明,相对于目前常见标准非负矩阵分解图像融合算法,采用区域突变度作为目标函数的动态加权非负矩阵分解算法平均梯度提高了36%以上,标准差提高了17%以上。

The image fusion algorithm based on standard non-negative matrix factorization has limited global feature extraction ability, resulting in the low contrast and poor visual effect of the fusion result. In order to improve the fusion effect of non-negative matrix factorization, a novel image fusion algorithm for infrared and visible images based on dynamic weighted non-negative matrix factorization is proposed. The weighted coefficients are designed for emphasizing the important characteristics of the source images, and they are modified after every iterative according to the relative importance variation of different areas, so the feature extraction ability of weighted non-negative matrix factorization is enhanced obviously. Compared with standard non-negative matrix factorization based fusion algorithm, proposed algorithm improves the visual effect, and the average gradient of the fusion result improves more than 36% and the standard deviation improves more than 17%.