最小生成树分割下小样本图像纹理提取研究
Research on Texture Extraction of Small Sample Images Based on Minimum Spanning Tree Segmentation
王智军 1郭艳光 2王鹏1
作者信息
- 1. 赤峰学院数学与计算机科学学院,内蒙古 赤峰 024000
- 2. 内蒙古农业大学计算机与信息工程学院,内蒙古 包头 014109
- 折叠
摘要
图像的纹理特征是图像的重要视觉特征,对于小样本图像的纹理特征提取时,存在纹理信息提取精度不佳、纹理信息提取错误等问题,严重影响了图像纹理提取的效果.为了有效解决以上问题,提出最小生成树分割下小样本图像纹理提取方法.采用Shearlet变换和多尺度Retinex方法对小样本图像实行增强处理,以提高其可识别性和区分度.利用最小生成树分割方法,对小样本图像分割处理;通过Gabor滤波器实现小样本图像的纹理提取.实验结果表明,所提方法能够有效地提取出小样本图像的纹理特征,其提取精度在97%以上,且图像增强效果佳.
Abstract
Texture features are important visual characteristics of images.In the extraction of texture features from small sample images,the issues such as low accuracy and big error during the texture information extraction have seri-ously affected the effectiveness of image texture extraction.To effectively solve these problems,a method for extracting texture features from small sample images was proposed based on minimum spanning tree segmentation.Firstly,Shear-let transform and multi-scale Retinex methods were adopted to enhance the small sample images to improve their rec-ognizability and distinguishability.Then,the minimum spanning tree segmentation method was used to segment the small sample images.Finally,the Gabor filter was used to extract the texture of small sample image.Experimental re-sults prove that the proposed method can effectively extract the texture features of small sample images with an extrac-tion accuracy of over 97%,and the image enhancement effect is excellent.
关键词
小样本图像/图像增强/最小生成树/滤波器/纹理特征提取Key words
Small sample image/Image enhancement/Minimum spanning tree/Filter/Texture feature extrac-tion引用本文复制引用
出版年
2024
NETL
NSTL
万方数据