基於二維EMD和遺傳演算法的多聚焦圖像融合

A New Multi-Focus Image Fusion Based on Two-Dimensional EMD and Genetic Algorithm

本文提出了一種基於二維經驗模態分解 （EMD） 和遺傳演算法的多聚焦圖像融合方法。源圖像通過二維經驗模態分解得到本征模函數 （IMF） 分量，根據T 檢驗對得到的本征模函數分量進行高低頻選擇。低頻係數融合使用改進的視窗資訊熵最高準則進行，高頻係數採用視窗相關性計算準則，視窗相關性係數融合根據閾值大小進行，閾值通過遺傳演算法確定。融合結果利用對融合係數採用二維經驗模態分解逆變換得到。模擬結果表明該演算法顯著優於基於圖元和基於視窗及基於小波的圖像融合方法。

A multi-focus image fusion method based on two-dimensionalempirical mode decomposition and genetic algorithm is presented in thispaper. First, a two-dimensional empirical mode decomposition is appliedto the decomposition of source images. High and low frequency ofintrinsic mode function component are classified by a T-test. Then lowfrequency coefficients are fused by improved maximum regionalinformation entropy criterion whereas the high frequency coefficients areamalgamated in different threshold ranges of coefficients by regionalcorrelation. The regional correlation threshold is selected by search ofgenetic algorithm. Finally, combined results are obtained by inversetwo-dimensional empirical mode decomposition transform on fusioncoefficients. Simulation results show that the proposed algorithmsignificantly outperforms traditional image fusion methods that are basedon the pixel, region, and wavelet, respectively.