基于非下采样剪切波变换与显著信息加权的图像融合算法

An Image Fusion Algorithm Based on Nonsubsampled Shearlet Transform Coupled with Significant Information Weighting

为了克服当前较多图像融合方法主要依靠测量图像能量信息来完成不同系数的融合,忽略了图像的显著内容,导致融合图像含有吉布斯效应及间断效应等弊端,设计了非下采样剪切波变换(nonsubsampled Shearlet transform,NSST)耦合显著信息加权的图像融合算法。引入NSST机制,对源图像进行系数分解,获取高、低频系数。借助高斯滤波器来构造出显著度量模型,以计算图像拥有的显著信息。随后,利用信息熵函数来计算出图像拥有的细节丰富度。并以图像拥有的细节丰富度和显著信息为依据,设计低频系数融合的加权因子,以此完成低频像素的融合。最后,利用图像中像素点的三邻点像素值,融合高频系数,获取融合图像。实验结果显示,与当前图像融合技术相比,所提算法融合质量更好,融合结果连续性较强,所对应的平均梯度值较大。

In order to overcome the disadvantages of Gibbs effect and discontinuity effect induced by relying on the measurement of image energy information to complete the fusion of different coefficients and ignoring the significant content of the image in current image fusion methods,an image fusion algorithm based on the nonsubsampled Shearlet transform(NSST)coupled with significant information weighting was proposed.The NSST mechanism was introduced to decompose the coefficients of the source image for obtaining high and low frequency coefficients.And the Gaussian filter was used to construct saliency measurement model for calculating the saliency information of image.Then,the information entropy function was used to calculate the detail richness of the image.Based on the detail richness and salient information of the image,the weighting factor of low-frequency coefficient fusion was designed to complete the fusion of low-frequency pixels.Finally,the fused image was obtained by fusing the high frequency coefficients with the pixel values of the three adjacent points in the image.Experimental results show that the proposed algorithm has better fusion quality,stronger continuity of fusion results and larger average gradient value compared with the current image fusion technology.