CAJ | 학술논문

针对存在的频谱混叠及图像细节信息丢失问题, 研究已有基于多尺度分解的多分辨率图像融合方法, 提出采用具有移不变性的非子采样轮廓变换对图像进行多尺度、多方向分解, 更完整地表征图像的几何结构, 同时对滤波器组进行上采样消除频谱混叠。研究已有的子带系数的融合规则, 提出了基于方向向量约束对比度的高频系数融合规则, 以及基于局部多向方差能量的选择与平均的低频系数融合规则。经多组实验结果表明, 该融合方法在视觉上和评价测度指标上都优于已有基于多分辨率图像融合方法。
침대존재적빈보혼첩급도상세절신식주실문제, 연구이유기우다척도분해적다분변솔도상융합방법, 제출채용구유이불변성적비자채양륜곽변환대도상진행다척도、다방향분해, 경완정지표정도상적궤하결구, 동시대려파기조진행상채양소제빈보혼첩。연구이유적자대계수적융합규칙, 제출료기우방향향량약속대비도적고빈계수융합규칙, 이급기우국부다향방차능량적선택여평균적저빈계수융합규칙。경다조실험결과표명, 해융합방법재시각상화평개측도지표상도우우이유기우다분변솔도상융합방법。
This paper investigated the existence problems of spectrum aliasing and detail lost and proposed a new novel fusion method based on the nonsubsmapled contourlet transform(NSCT). It adopted the new decomposition of NSCT, which built upon nonsubsampled pyramids and nonsubsampled directional filter banks for extracting the geometric information of images. It proposed the high-frequency coefficient fusion rules by combination of direction vector and contrast, and proposed the low-frequency fusion rule based on the local energy of multi-direction variance. Experimental results show that the proposed method achieves</sup> better fusion results than wavelet-based and CT-based fusion method.