结合SML与差分图像的多聚焦图像融合算法

Multi-focus image fusion algorithm based on SML and difference image

针对传统多聚焦图像融合算法中融合边缘出现模糊、伪影等问题,提出了一种结合改进拉普拉斯能量和(SML)与差分图像的多聚焦图像融合算法。首先,为了提取源图像的聚焦特征信息,分别通过SML和滤波差分进行聚焦度量,再采用引导滤波获得更多的细节特征;接着,利用像素最大值规则生成初始融合决策图,再对初始融合决策图进行小区域去除消除因聚焦和散焦区域相似造成的噪点,并对融合决策图进行不一致处理,获得更精确的聚焦区域;最后,由逐像素加权平均规则,得到融合图像。实验结果表明,所提出的算法在主观视觉效果和客观评价指标上均优于对比算法,互信息、特征互信息、图像梯度特征在彩色图像上分别提高了0.17%、0.38%和0.11%,在灰度图像上分别提高了0.7%、0.69%和0.33%,并且平均运行时间少于0.5 s,具有较高的计算效率。此外,该算法能够较好地保留源图像信息的完整性,融合图像边缘清晰、无伪影。

Aiming at the problems of edge blurring and artifacts in traditional multi-focus image fusion algorithm,a multi-focus image fusion algorithm based on sum-modified-Laplacian(SML)and difference image is proposed.Firstly,in order to extract the focus feature information of source images,the focus measurement is performed by SML and difference operation between filtered images and source images respectively,and the guided filtering is adopted to obtain more detailed features.Then,the initial fusion decision map is generated by using pixel-wise maximum rule,the small area removal strategy is performed to eliminate the noise caused by the similarity of focus and defocus areas,and inconsistency processing is used to generated final decision map to obtain a more accurate focus region.Finally,the fused image is obtained by the pixel-by-pixel weighted average rule.Experimental results indicate that the proposed algorithm is superior to the comparison algorithm in both subjective visual effect and objective evaluation metrics.The mutual information,feature mutual information and image gradient features are improved by 0.17%,0.38%and 0.11%respectively on color images.On grayscale images,the improvements are 0.7%,0.86%and 0.47%,respectively.Furthermore,the average consuming time is less than 0.5 s with high computational efficiency.In addition,the algorithm can better retain the integrity of source image information,and the fusion image has clear edges without artifacts.