Fusion of Infrared Polarization and Intensity Images Based on Two Dimension Variational Mode Decomposition

Abstract

This paper presents a new fusion algorithm by using infrared polarization and intensity images based on two dimensional variation mode decomposition (2D-VMD). It effectively extracts features from both infrared polarization and intensity images, and efficiently reduces information loss in the fusion process. The fusion algorithm firstly decomposes each of the two source images into a basic image and a set of intrinsic mode functions (IMF) through the 2D-VMD. These IMFs have specific directional and oscillatory characteristics and are the high frequency components of a source image. Secondly the basic image is further decomposed into a contour feature image and a texture image by using robust principal component analysis (RPCA).Thirdly the two contour feature images from the polarization basic image and intensity basic image are integrated by applying arctangent contrast modulation, the texture two images are combined based on a method, namely local average gradient summation, and then the fused basic image is obtained by the inverse transformation of RPCA. Fourthly, each high frequency component is fused by using the maximum rule. With both the fused basic image and high frequency sub-images, the final fused image is obtained by inverse transformation of 2D-VMD. The evaluation is conducted based on the comparison of the proposed fusion algorithm with other five existing fusion algorithms.