一种快速的基于稀疏表示和非下采样轮廓波变换的图像融合算法

为了提高图像融合的效率和质量，该文提出一种基于快速非下采样轮廓波变换(NSCT)和4方向稀疏表示的图像融合算法。该方法首先对源图像进行快速NSCT分解，生成一系列低通和高通子带。对于低频子带，利用自适应生成的DCT过完备字典进行快速的4方向稀疏表示和系数融合；对于高频子带，则利用高斯加权区域能量最大的融合规则进行系数融合。快速NSCT将传统NSCT的树形滤波结构转变为多通道滤波结构，能成倍提高分解效率；快速的稀疏融合则抛弃了传统的滑动窗口方法，以水平、垂直、对角线4个方向进行稀疏表示和稀疏融合，进一步提高算法效率。实验结果表明，提出的快速算法能在不影响融合质量的条件下将算法效率提高近20倍。

Fast Image Fusion Algorithm Based on Sparse Representation and Non-subsampled Contourlet Transform

In order to improve the efficiency and quality of image fusion, a new image fusion algorithm based on four-direction Sparse Representation (SR) and fast Non-Subsampled Contourlet Transform (NSCT) is proposed. The proposed method firstly provides a series of low- and high-frequency sub-bands of source images via fast NSCT decomposition. Then adaptive DCT over-complete dictionary is used for the fast four-direction sparse representation and coefficients fusion of low-pass sub-band, while Gaussian weighted regional energy based fusion rule are used in high-pass sub-bands. Fast NSCT modifies the tree structure filter bank of traditional NSCT into multi-channel structure, and it saves about half of the time. The fast SR fusion method adopts a four-direction sparse representation for coefficients fusion instead of traditional sliding window method, and further improves the efficiency of algorithm. The experimental results show that the proposed fast fusion algorithm can improve the efficiency nearly 20 times without sacrificing fusion quality.