多聚焦图像融合技术是将同一场景下多幅聚焦区域不同的图像进行整合，生成一幅清晰度更高且信息内容更丰富的全聚焦图像，以此提高信息的利用率。但是现有的多聚焦图像融合算法无法准确地判别图像边界区域的聚焦属性，而且不能有效地提取较小的聚焦区域，易导致融合图像存在伪边缘、块效应、小区域模糊的现象。为此，本文基于多聚焦图像成像特性，提出了改进的多聚焦图像融合方法，主要内容如下：

（1）提出了一种基于显著特征提取和焦点检测的图像融合算法。针对传统的焦点检测方法无法准确判断图像中均匀区域的聚焦属性，易导致融合图像存在伪边缘和块效应等问题，提出了一种加权局部方差焦点检测方法。该方法结合高斯曲率滤波器在显著特征提取方面的优势，在局部窗口内对显著特征图的局部方差赋权值求和，增加了周围局部方差强度对该区域的贡献；接着，对焦点检测结果进行最大化准则判决生成决策图，并采用孔洞移除策略和引导滤波器进行优化；最后，利用优化后的决策图对源图像加权融合得到融合图像。实验结果表明，所提算法具有较好的聚焦区域检测能力，能够保留多聚焦图像丰富的细节信息，避免融合图像存在块效应和伪影现象。

（2）提出了一种基于多尺度焦点检测和随机游走的图像融合算法。针对目前众多空间域算法无法精确检测小区域的问题。本文采用多尺度加权局部方差获取显著特征图在不同尺度的焦点检测结果，以充分考虑不同大小的邻域内像素间的相关性，并由此构建多尺度决策图；此外，在随机游走模型中通过自定义范围确定阈值的方法，对不同尺度的决策图具有针对性的求解权重，以结合多尺度决策图的互补优势得到最终决策图；最后，根据所得决策图对源图像加权融合构建融合图像。实验结果表明，该方法能有效提高源图像中小区域的检测精度，完整地保留图像的聚焦信息，获得的融合图像空间一致性较高，在复杂场景下具有良好的融合效果。

Multi-focus image fusion technology integrates multiple images with different focus areas in the same scene to generate a fully focused image with higher definition and richer information content, so as to improve the utilization rate of information. However, the existing multi-focus image fusion algorithm cannot accurately determine the focusing properties of the image boundary area, and cannot effectively extract the small focus area, which can easily lead to the phenomenon of pseudo-edge, block effect, and small area blurring in the fused image. Therefore, based on the imaging characteristics of multi-focus images, this paper proposes improved multi-focus image fusion methods, the main contents are as follows:

(1)An image fusion algorithm based on saliency feature extraction and focus detection is proposed. Aiming at the problems that the traditional focus detection method cannot accurately judge the focus attribute of the uniform area in the image, which is easy to lead to the pseudo-edge and block effect of the fused image, a weighted local variance focus detection method is proposed. Combined with the advantages of Gaussian curvature filter in salient feature extraction, this method summes the local variance weights of the salient feature map in the local window, and increases the contribution of the surrounding local variance intensity to the region; Then, the decision graph is generated by maximizing the focus detection results, and the hole removal strategy and guide filter are used to optimize. Finally, the optimized decision graph is used to weighted fuse the source image to obtain the fused image. Experimental results show that the proposed algorithm has good focus area detection ability, can retain the rich detail information of multi-focus images, and avoid block effects and artifacts in fusion images.

(2)An image fusion algorithm based on multi-scale focus detection and random walk is proposed. Aiming at the problem that many spatial domain algorithms cannot accurately detect small regions. In this paper, multi-scale weighted local variance is used to obtain the focal point detection results of salient feature maps at different scales, so as to fully consider the correlation between pixels in neighborhoods of different sizes, and construct multi-scale decision maps; In addition, in the random walk model, the method of determining the threshold by custom range is used to solve the decision maps of different scales with targeted weights, so as to combine the complementary advantages of multi-scale decision charts to obtain the final decision map; Finally, the source image is weighted and fused according to the obtained decision graph. The experimental results show that this method can effectively improve the detection accuracy of small areas in the source image, completely retain the focusing information of the image, obtain high spatial consistency of the fusion image, and have a good fusion effect in complex scenes.