In this study, we propose a new method for multifocus image fusion by combining with the structure tensors of mixed order differentials and the multiscale neighborhood. In this method, the structure tensor of an integral differential is utilized to detect the high frequency regions and the structure tensor of the fractional differential is used to detect the low frequency regions. To improve the performance of the fusion method, we propose a new focus measure based on the multiscale neighborhood technique to generate the initial fusion decision maps by exploiting the advantages of different scales. Next, based on the multiscale neighborhood technique, a post-processing method is used to update the initial fusion decision maps. During the fusion process, the pixels located in the focused inner regions are selected to produce the fused image. In order to avoid discontinuities in the transition zone between the focused and defocused regions, we propose a new “averaging” scheme based on the fusion decision maps at different scales. Our experimental results demonstrate that the proposed method outperformed the conventional multifocus image fusion methods in terms of both their subjective and objective quality.

在这项研究中，我们提出了一种新的方法，结合混合阶差和多尺度邻域的结构张量，进行多焦点图像融合。在这种方法中，积分微分的结构张量用于检测高频区域，分数微分的结构张量用于检测低频区域。为了提高融合方法的性能，我们提出了一种基于多尺度邻域技术的新的聚焦度量，以利用不同尺度的优势来生成初始融合决策图。接下来，基于多尺度邻域技术，使用后处理方法来更新初始融合决策图。在融合过程中，选择位于聚焦内部区域中的像素以产生融合图像。为了避免在聚焦区域和散焦区域之间的过渡区域中出现不连续性，我们基于不同尺度上的融合决策图提出了一种新的“平均”方案。我们的实验结果表明，该方法在主观和客观质量上均优于传统的多焦点图像融合方法。