W变换和NSCT相结合的多聚焦图像融合方法

发布时间：2019-01-07 20:39

【摘要】：图像的多尺度分解技术和融合规则是决定多聚焦图像融合效果的关键因素。k次W系是一类以k次多项式和k次分段多项式为基函数的新的混合正交函数系统,对应的W变换是一种具有正交性和精确重构性的有效多分辨分析工具。结合W变换的多尺度特点和非子采样方向滤波器组变换的多方向性,提出了一种新的基于W变换和非子采样方向滤波器组(NSDFB)的多尺度多方向变换。该变换利用W变换对图像进行多尺度分解,利用二维NSDFB对W分解的高频子带系数进行方向分解,得到不同尺度不同方向的子带图像。在此基础上,提出了一种新的多聚焦图像融合算法。该算法针对多聚焦图像高频系数的特点,改进了常用简化脉冲耦合神经网络算法,并将其用于高频系数的融合规则中。实验结果表明,提出的融合方法能够有效地选择源图像中的聚焦良好区域,抑制伪影信息,产生视觉效果更好的融合图像,且在标准差、信息熵、平均梯度和空间频率等客观评价指标上都优于传统的基于Contourlet变换、非下采样Contourlet变换、离散小波变换的融合方法。
[Abstract]:The multi-scale decomposition technique and fusion rules are the key factors to determine the fusion effect of multi-focus image. The k-degree W system is a new hybrid orthogonal function system with the k-degree polynomial and the k-degree piecewise polynomial as the basis functions. The corresponding W transform is an effective multi-resolution analysis tool with orthogonality and precision reconstruction. Based on the multi-scale characteristic of W transform and the multi-directivity of non-subsampling directional filter bank transform, a new multi-scale multi-directional transformation based on W transform and (NSDFB) is proposed. The transform uses W transform to decompose the image in multi-scale, and two-dimensional NSDFB to decompose the high-frequency subband coefficients of W decomposition in the direction, and obtain the sub-band images with different scales and different directions. On this basis, a new multi-focus image fusion algorithm is proposed. According to the characteristics of high frequency coefficients of multi-focus images, this algorithm improves the common simplified pulse coupling neural network algorithm, and applies it to the fusion rules of high frequency coefficients. The experimental results show that the proposed fusion method can effectively select a good focus area in the source image, suppress the artifacts, and produce a better fusion image with better visual effect, and at the standard deviation, the information entropy, The average gradient and spatial frequency are superior to the traditional fusion methods based on Contourlet transform, non-downsampling Contourlet transform and discrete wavelet transform.
【作者单位】：北京林业大学理学院;
【基金】：国家自然科学基金(61571046,61372190,61370193)资助项目
【分类号】：TP391.41

【参考文献】