适于生物细胞图像的多尺度混合分层融合算法

Hybrid Multi-Scale and Multi-Level Fusion Algorithm for Biological Cell Image

绿色荧光蛋白（GFP）荧光图像提供蛋白质功能定位信息，相衬图像提供高分辨率的结构信息，二者的融合对于细胞蛋白质的功能分析和亚细胞的精细结构定位具有重要价值。基于轮廓波变换对细节信息优异的表达能力，提出一种HIS空间下基于轮廓波变换的多尺度混合分层图像融合算法，该算法利用轮廓波变换分别对GFP荧光图像和相衬图像的亮度分量进行分解，对不同子带系数采用不同的融合策略以力求融合图像在保留GFP荧光图像定位信息的同时融入相衬图像的高频细节，并引入视觉保真度（VIF）作为图像融合的评价指标。对来自约翰英纳中心的117组拟南芥细胞图像进行的融合实验表明，该算法能够有效保留源图像中的细节信息，提高了融合图像的可视性，相比传统算法更为优越。

Green fluorescent protein （GFP） provides important localization information of protein while phase- contrast image provides the structure information of high resolution. The combination of GFP fluorescent image and phase-contrast image is valuable for function analyses of protein and adequate structure localization of subcellular. We propose a hybrid multiscale and multi-level image fusion algorithm which is based on HIS and contourlet transform, in order to take advantage of contourlet＇s ability of directional and excellent detail expression. This algorithm uses contourlet transform to decompose the intensity components of both GFP fluorescent image and phase-contrast image, and different fusion schemes are ultilized to different sub-band coefficients in order to keep the localization information of GFP fluorescent image and detail information of phase-contrast image in the fused image. Visual information fidelity （VIF） is introduced to evaluate the fusion result. The experimental results from 117 groups of arabidopsis cell images of John Innes Center show that the proposed algorithm can both neep the details of original images well preserved and improve the visibility of the fusion image, which shows the superiority of the novel method to traditional ones.