多模态图像局部极值点特征生成方法仿真  

Simulation of Local Extreme Point Feature Generation Method for Multimodal Images

作　　者：张镭赋 高家骥 ZHANG Lei-fu;GAO Jia-ji(Dalian Polytechnic University,Dalian Liaoning 116034,China)

机构地区：[1]大连工业大学,辽宁大连116034

出　　处：《计算机仿真》2024年第8期481-485,共5页Computer Simulation

基　　金：教育部产学合作协同育人项目,项目编号:23110324216382。

摘　　要：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。Aiming at the problem of difficulty in feature generation caused by the variety of multimodal image features and the presence of noise interference,a local image feature generation method based on improved CNN is proposed.Firstly,we introduced the attention mechanism to calculate the information modality correlation between the previous moment and the next moment in pixel distribution sequence.Secondly,we gave equal weights to different regions in the image,thus obtaining regional information.Thirdly,we constructed a gray co-occurrence matrix,where each layer has pixel values coresponding to itself.Then,we extracted the global gray-level mean of the image to be generated.In the matrix,we searched the corresponding level of local extreme points.Meanwhile,we extracted the gradient magnitude and partial derivative coefficient in the level.Finally,we debugged and compared the partial derivative coefficient with the global feature,thus achieving the local feature generation of multimodal image.The experimental results show that the proposed method has good effects in feature generation of sample images,and it is basically unaffected by image noise and local occlusion.

关 键 词：改进CNN 多模态图像 局部特征生成 全局灰度均值 偏导系数 灰度共生矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]