压缩感知在多源图像融合中的应用

Application of compressed sensing in multi-source image fusion

摘要:

针对全采样传统图像融合方法中计算量大、时间复杂度高的问题，提出了一种基于压缩感知(CS)理论的多源图像融合模型。为满足一定的稀疏性，将源图像在过完备二维离散余弦变换(DCT)字典上进行稀疏表示，并通过随机观测得到待融合的观测值；在每一图像块上采用基于标准差的方法自适应地计算融合权值，加权合成融合后的观测值，然后利用改进步长的梯度追踪算法求解稀疏系数，得到最终融合图像。实验结果表明：与传统方法相比，提出的融合模型在减少计算量和存储容量的同时，能更好地从源图像中提取信息，获得效果较好的融合图像。

Abstract:

A new fusion method based on Compressed Sensing(CS) is proposed to solve storage and computation cost problems in traditional image fusion algorithms. Sparse representation coefficients of source images are obtained on the basis of overcomplete two-dimensional Discrete Cosine Transform(DCT) dictionary. Then the observed values which will be fused are got by applying random projection on the coefficients. The weights of each image block are calculated adaptively based on standard deviation method. Thus input image measurements are fused into composite measurements via weighted averaging. The fused image is reconstructed through improved gradient pursuit with modified stepsize. The simulation results show that, comparing with other fusion algorithms, the proposed method can achieve better performance on fusion results with less sampling numbers and low computational complexity.