多模态图像局部极值点特征生成方法仿真

针对因多模态图像特征种类较多、存在噪声干扰导致的特征生成难度较大的问题,提出一种基于改进CNN的图像局部特征生成方法。引入注意力机制,计算像素分布序列中上一时刻与下一时刻之间的信息模态关联,针对图像中的不同区域均给出同等权重,获取区域信息。建立灰度共生矩阵,矩阵中每一层都有与自身相对应的像素值,提取待生成图像的全局灰度均值,在矩阵中查找局部极值点的对应层次,提取该层次中梯度量级和偏导系数,通过偏导系数与全局特征调试对比,实现多模态图像局部特征生成。实验结果表明,所提方法针对样本图像的特征生成效果较好,基本不受图像噪点和局部遮挡的影响。

Domain-based noise removal method using fourth-order partial differential equation

Due to the fact that the fourth-order partial differential equation （PDE） for noise removal can provide a good trade-off between noise removal and edge preservation and avoid blocky effects often caused by the second-order PDE, a domain-based fourth-order PDE method for noise removal is proposed. First, the proposed method segments the image domain into two domains, a speckle domain and a non-speckle domain, based on the statistical properties of isolated speckles in the Laplacian domain. Then, depending on the domain type, different conductance coefficients in the proposed fourth-order PDE are adopted. Moreover, the frequency approach is used to determine the optimum iteration stopping time. Compared with the existing fourth-order PDEs, the proposed fourth-order PDE can remove isolated speckles and keeps the edges from being blurred. Experimental results show the effectiveness of the proposed method.

基于精确Zoeppritz方程的叠前三参数反演

本文提出基于精确Zoeppritz方程的叠前三参数反演,推导了精确解的反射系数偏导,从而构建用于反演的雅可比矩阵;比较了精确Zoeppritz方程和Aki-Richards近似方程的反演精度和稳定性,建立了基于广义线性反演的叠前反演方法。