基于块分解的超分辨率图像特征零样本分类  

Zero-Sample Classification of Super-Resolution Image Features Based on Block Decomposition

作　　者：张杰 汤嘉立 高伟 ZHANG Jie;TANG Jia-li;GAO Wei(School of Computer Engineering,Jiangsu University of Technology,Changzhou Jiangsu 213000,China;School of Electrical Engineering,Southeast University,Nanjing Jiangsu 210096,China)

机构地区：[1]江苏理工学院计算机工程学院,江苏常州213000 [2]东南大学电气工程学院,江苏南京210096

出　　处：《计算机仿真》2022年第7期248-251,262,共5页Computer Simulation

基　　金：常州市科技支撑计划(社会发展)项目(CE20215029);江苏高校哲学社会科学研究一般项目(2019SJA1061,2020SJA1173);江苏理工学院教学改革与研究项目(11610312033,11610311934)。

摘　　要：当前的图像特征零样本分类方法忽略了图像块的分解,导致特征分类结果误差较大、真实性较差。为此提出基于块分解的超分辨率图像特征零样本分类方法。利用块分解方法将图像分解为图像块,采用小波分解方法完成图像去噪。构建图像语义融合模型,融合图像中语义特征。利用自适应加权算法分类语义特征,实现图像特征零样本分类。实验结果表明,图像零样本分类方法的准确度较高,AUC指标测试结果验证了研究方法的分类结果具有理想的真实性。Currently,some zero-sample classification methods ignore the decomposition of image blocks,leading to large error and low authenticity of feature classification results.Therefore,a method of zero-sample classification for super-resolution image features based on block decomposition was put forward.First of all,the block decomposition method was used to divide the image into some blocks,and then the wavelet decomposition method was used to remove the noise form images.Moreover,the model of image semantic fusion was constructed to fuse the semantic features in images.Finally,the self-adaptive weighted algorithm was used to classify semantic features and thus to achieve the zero sample classification for image features.Experimental results show that the accuracy of the proposed method is higher than before.The AUC index test results prove that the classification results are ideal and real.

关 键 词：块分解 超分辨率图像 特征提取 零样本分类 小波包变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]