语义兴趣点检测算法在图像分割中的研究  

Research on semantic interest points detection algorithm in images segmentation

作　　者：陆宏菊[1] LU Hongju(School of Information Engineering,Jinan Technician College,Jinan 250115,China)

机构地区：[1]济南市技师学院信息工程学院,济南250115

出　　处：《智能计算机与应用》2022年第11期87-91,共5页Intelligent Computer and Applications

基　　金：国家自然科学基金(61902225)。

摘　　要：针对不同类型的图像,不同模态的特征能够对图像分割提供不同的帮助。本文提出一种基于无监督学习策略的多模态特征映射模型,通过对多模态特征在解空间的投影将图像分割问题转化为最小化像素相似度距离的优化问题。本文提出算法通过无监督学习策略得到特征映射矩阵,实现将高维度的多模态特征进行投影得到低维的特征解平面,用来进行图像的前景-背景分割。最后,在BSD500数据集中进行了对比,验证了本文算法的有效性和先进性。For different types of images, the features of different modalities can provide different help for images segmentation.In this paper, a multimodal feature mapping model based on the unsupervised learning strategy is proposed, which transforms the image segmentation problem into an optimization problem of minimizing the pixel similarity distance by projecting the multimodal features into the solution space.The algorithm in this paper obtains the feature mapping matrix through the unsupervised learning strategy, and realizes the projection of the high-dimensional multimodal features to obtain the low-dimensional feature solution plane for the foreground-background segmentation of the image.Finally, a comparison is made in the BSD500 dataset to verify the effectiveness and advancement of the algorithm proposed in this paper.

关 键 词：图像自动分割 多模态特征 特征映射 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]