摘要
针对Trace变换提取的图像特征缺乏对纹理边缘信息描述和计算代价高的问题,利用小波变换对图像轮廓的表征优势,提出了多分辨率Trace变换并应用于纹理图像分类.首先,将小波变换引入到Trace变换中,对纹理图像进行非下采样小波变换,得到不同频率的低频特征子图及高频边缘子图;其次,在各级子图上进行一组泛函的Trace变换,获取纹理图像的融合特征,在获得图像边缘信息的同时避免了Trace变换不同泛函组合计算代价过高的问题;最后,把融合特征送入支持向量机对图像进行分类.实验结果表明,对图像采用多分辨率Trace变换提取的融合特征具有更好的纹理描述能力,相对于传统Trace变换及MCM等对比方法具有更高的鉴别性能,且在时间效率上相对于传统Trace变换有大幅提升.
There is a problem that the image features extracted by trace transform lack description of texture edge information,and the computation cost is high,too.Based on the advantages of wavelet transform in image contour representation,a new fusion feature extraction algorithm,multi-resolution trace transform,is proposed and applied to texture image classification.Firstly,the wavelet transform was introduced in trace transform,low frequency feature sub images and high frequency edge sub images of texture images at different frequencies are obtained by using nonsubsampled wavelet transform.Then,we carried out a set of functional trace transform on each level sub images to obtain the fusion features of texture image,which not only obtains the edge information of the image,but also avoids the problem of high cost.Finally,the fusion features were fed into support vector machines to classify the images.The experiment results show that the fusion features of multi-resolution trace transformation have better texture description ability and achieves higher recognition rate than the original trace transform and MCM contrast method,the time efficiency is greatly improved compared to the traditional trace transform.
作者
黎明
邢冬冬
汪宇玲
LI Ming;XING Dong-dong;WANG Yu-ling(Key Laboratory of Jiangxi Province for Image Processing and Pattern Recognition,Nanchang Hangkong University,Nanchang,Jiangxi 330063,China;Jiangxi Engineering Laboratory on Radioactive Geoscience and Big Data Technology,East China University of Technology,Nanchang,Jiangxi 330013,China;School of Information Engineering,Nanchang Hangkong University,Nanchang,Jiangxi 330063,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2019年第4期962-969,共8页
Acta Electronica Sinica
基金
国家自然科学基金(No.61866025)
江西省优势创新团队(No.20113BCB24009
No.20181BCB24008)
江西省教育厅科技项目(No.GJJ170432)
江西省图像处理与模式识别重点实验室开放基金(No.ET201880042)
江西省放射性地学大数据技术工程实验室开放基金(No.JELRGBDT201804)
关键词
纹理分类
Trace变换
非下采样小波变换
多分辨率
texture classification
trace transform
nonsubsampled wavelet transform
multi-resolution