Color image fusion method using the multi-scale retinex and directional support value transform

Abstract

Completely different from many exiting multi-source image fusion methods, a novel color image fusion method for a single original color image, combining the multi-scale Retinex (MSR) with directional support value transform (DSVT), is presented in this paper. The applied MSR is an (centre/surround-based) Retinex algorithm and the directional support value transform, an anisotropic and multi-scale transform, is deduced under the weighted mapping least squares support vector machine (LS-SVM) framework. Using the MSR to an original color image, an enhanced color image, producing more detail information hidden in shadow areas of original color image but easily lapsing into color distortion, is obtained. In the HSV color space, the good color information of original color image and more detail information of enhanced color image are integrated into a fused color image based on the DSVT. Series of color images under different environment are chosen for color image fusion experiments and the performance of DSVT is compared with other methods also used in color image fusion, including Laplacian pyramid, discrete wavelet transform and support value transform. The experimental results demonstrate that: the proposed color image fusion approach is effectively making the fused color images not only present more clearly detail information but also maintain color fidelity; DSVT is superior to other three methods used in color image fusion.