Abstract: The existing Empirical Mode Decomposition（EMD） methods usually ignore the correlation of each channel image when decomposing a multi-channel image（e.g. a color image）. To ameliorate, this paper presents a novel EMD method for multi-channel images. It computes the upper and lower envelopes of multi-channel images by an interpolation approach based on bi-Laplacian operator, and sets up a whole stopping criterion of the sifting process to consider the correlation of all channel images. The novel EMD method can decompose a multi-channel image into a finite number of Intrinsic Mode Functions（IMFs） with different scale features and a residue representing the whole change trend of the image. According to the novel EMD method, a suite of challenging application tasks in image analysis and processing can be undertook, such as image sharpening and night image enhancement. Experimental results demonstrateal this method and its applications can generate good results.

Key words: multi-channel images, empirical mode decomposition, bi-Laplacian operator, image analysis and processing

摘要： 针对现有的经验模态分解方法（Empirical Mode Decomposition，EMD）对多通道图像（如彩色图像）进行分解时通常忽略各通道图像之间相关性的问题，提出了一种多通道图像EMD方法。该方法采用双拉普拉斯算子插值得到图像上下包络，并建立一个整体筛分停止准则进行筛分来考虑各通道图像相关性，能够将多通道图像自适应分解为数目不多的内蕴模态函数（Intrinsic Mode Function，IMF）分量和一个余量，其中内蕴模态函数分量体现了原始图像不同尺度的特征信息，余量体现了图像的整体变化趋势。该方法可以应用在图像锐化、夜景图像增强等图像分析和处理领域。实验结果显示该方法能够取得较好的效果。

关键词: 多通道图像, 经验模态分解, 双拉普拉斯算子, 图像分析和处理