基于粒子群优化的多分辨率图像融合  

Multiresolution Image Fusion Using Particle Swarm Optimization

作　　者：赵诚[1] 付梦印[1] 王立平[1] 

机构地区：[1]北京理工大学自动化学院,北京100081

出　　处：《兵工学报》2010年第2期171-176,共6页Acta Armamentarii

基　　金：国家自然科学基金资助项目(60773044)

摘　　要：提出了一种基于粒子群优化的多分辨率图像融合算法,用以融合红外与可见光视觉传感器获取的图像。分别对原始图像执行快速离散Curvelet变换;根据不同子带系数的特性与原始图像的光谱特征,在低频系数的融合中着重保留目标特征,并对其余系数选取基于Tsallis熵的互信息量作为评价指标,进而利用改进的粒子群优化算法求取最佳系数融合权值;对各高频子带系数采用基于局部区域能量匹配的融合规则。经过Curvelet逆变换得到融合结果图像。实验结果表明,该算法可以有效地综合红外图像中的目标特征与可见光图像中的细节信息,其融合结果在主观视觉效果与客观评价指标上均优于传统的基于塔形变换与小波变换的融合算法。In order to fuse images acquired by infrared and visible imaging sensors, a novel muhiresolution image fusion algorithm based on PSO was proposed. It is basic steps of the image fusion algorithm that the fast discrete Curvelet transform is implemented on source images, respectively; for the low-frequency subband, coefficients of infrared targets are preserved to retain maximally target features, and best fusion weights of other coefficients are determined by taking mutual information defined with the Tsallis entropy as the evaluation index and using PSO; for various high-frequency subbands, a fusion rule based on local energy matching is chosen; the fused resutt is obtained by implementing the inverse Curvelet transform. Ths simulated results show that the proposed algorithm can effectively integrate the target features and details of the infrared and visible images into the fused results, respectively, and has better performance than traditional pyramid-based and wavelet-based fusion algorithms both in subjective visual quality and objective evaluation indices.

关 键 词：信息处理技术 图像融合 CURVELET变换 粒子群优化 TSALLIS熵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]