基于区域特性的压缩感知多聚焦融合算法

Multi-focus Image Fusion Algorithm Based on Compressed Sensing and Regional Characteristics

传统的基于压缩感知的图像融合算法是对整个系数进行稀疏处理,而小波分解后的低频系数不稀疏,导致压缩重构质量降低,并且传统的融合规则不易简单、全面地提取高频系数的特征值。针对这一问题,分别对小波分解得到的高、低频系数采取不同的融合规则进行处理,提出了一种改进的区域特性高频压缩感知的融合算法。其中,低频系数采用区域方差加权绝对值最大融合;高频系数首先通过具有较好RIP性质的随机观测矩阵进行压缩采样,得到的观测值基于能量匹配度的不同进行相加或加权融合,以融合不同方向的高频子带特征信息,再用正交匹配追踪重构算法对高频部分进行信号重构。最后,低频、高频信息在小波逆变换下重构出融合图像。实验结果表明,与以往的基于压缩感知的融合方法相比,此算法的融合图像更清晰,新算法无论是在主观评价还是客观评价指标上都有利于图像信号重构,并具有较好的使用性。

The traditional image fusion based on compressed sensing algorithm is to deal with the whole coefficients sparse, and low-frequency coefficients of wavelet decomposition is not sparse, resulting in the quality reduction of com- pression reconstruction. Besides, the traditional fusion rules are difficult and not comprehensive to extrat characteristic value of high frequency coefficient. To solve this problem, we dealt with high and low frequency coefficient which was decomposed by wavelet by adopting different fusion rules, and an improved fusion method based on high-frequency com- pressed sensing of regional characteristics was proposed. Among them, low-frequency coefficients fusion method used the regional variance of weighted and maximum absolute value. Firstly, the high-frequency coefficients by random meas- urement matrix has better restricted isometry property compression sampling. The observed value based on energy matching degree is used for different additive or weighted fusion, to fuse the characteristic information of high frequency sub bands in different directions. Then the orthogonal matching pursuit recovery algorithm is used to to reconstruct the signal of high-frequency part. Finally, the low-frequency and high-frequency information in invert wavelet transform are used for reconstructing the fusion image. Experimental results show that compared with the previous fusion method based on compressed sensing, the effect of the fused image is more clear, new algorithm both in subiective evaluation and objective evaluation index are conducive to the image signal reconstruction,and has good usability.