参数自适应脉冲耦合神经网络的医学图像融合算法

Medical image fusion based on parametric adaptive pulse coupled neural network

针对当前大多数传统的多模态医学图像融合技术不能同时实现能量保存和细节提取而导致信息不足和细节模糊等问题,提出一种基于参数自适应脉冲耦合神经网络(pulse coupled neural network,PCNN)的医学图像融合算法。首先,从标准库里取出经过配准的四种不同类型的源图像进行非下采样剪切波(non-subsampled shearlet transform,NSST)分解,得到多尺度多方向的低频子带系数和一系列高频子带系数;接着,采用参数自适应PCNN对高频子带进行融合,增强后的高频子带绝对值作为反馈输入自适应调节所有的PCNN参数。低频子带采用改进的平均梯度与空间频率相结合的策略进行图像融合,可以同时做到能量保存与细节提取;最后,经NSST逆变换得到融合后的医学图像。分别对灰度一灰度图像和灰度一彩色图像进行大量的算法融合实验,选取五种具有代表性的医学图像融合算法与之相比,并选择信息熵(IE)、标准差(SD)、边缘强度(ES)、平均梯度(AVG)、融合因子(FF)、结构相似度(SSIM)和清晰度(MG)七种指标对融合后的图像进行质量评价。仿真结果显示,与其它几种算法相比,该融合算法得到的信息熵、标准差、边缘强度、平均梯度、融合因子、结构相似度和清晰度的平均值分别提高了10.48%、2.86%、3.48%、4.92%、4.17%、4.19%和5.12%。融合所得的图像在视觉效果上也有较大优势。实验结果表明,与其他几种算法相比,本文算法在主观视觉以及客观数据上均优于其他算法,具有广阔的应用前景。

Aiming at the problem that most traditional multi-scale medical image fusion technologies cannot simultaneously save energy and extract details,which leads to insufficient information and blurred details,a medical image fusion algorithm based on parameter adaptive pulse coupled neural network(PCNN)is proposed Firstly,four different types of source images are extracted from the standard library for Non-subsampled shearlet transform(NSST)decomposition to obtain multi-scale and multi-directional low-frequency sub-band coefficients and a series of high-frequency sub-band coefficients;secondly,parameter adaptive PCNN is used to fuse high-frequency sub-segments,and the absolute value of enhanced high-frequency sub-segments is used as feedback input.Adapt to adjust all PCNN parameters.Low-frequency sub-segments uses improved average gradient and spatial frequency strategy to fuse images,which can save energy and extract details at the same time.Finally,the fused medical images are obtained by NSST inverse transformatioru Five representative medical image fusion algorithms are selected to compare with them.Seven indexes,namely information entropy(IE)standard deviation(SD),edge intensity(ES),average gradient(AVG),fusion factor(FF),structural similarity(SSIM)and clarity(MG),are selected to evaluate the quality of the fused image.The simulation results show that compared with other algorithms,the average information entropy,standard deviation,edge strength,average gradient,fusion factor,structural similarity and clarity of the fusion algorithm are improved by 10.48%,2.86%,3.48%,4.92%,4.17%,4.19%and 5.12%,respectively.The fusion image also has a great advantage in visual effect.The experimental results show that,compared with other algorithms,the proposed algorithm is superior to other algorithms in subjective vision and objective data,and has broad application prospects.