计算机工程

• 图形图像处理 • 上一篇    下一篇

自适应的纹理图像光照不变特征提取方法

尚赵伟1,胡德恒1,赵恒军1,杨 君1,2   

  1. (1. 重庆大学计算机学院,重庆 400030;2. 中国人民解放军西藏林芝军分区,西藏 林芝 860100)
  • 收稿日期:2012-07-16 出版日期:2013-11-15 发布日期:2013-11-13
  • 作者简介:尚赵伟(1968-),男,副教授、博士,主研方向:数字图像处理,模式识别;胡德恒,硕士研究生;赵恒军,博士研究生;杨 君,硕士研究生
  • 基金项目:
    国家自然科学基金资助项目(61173130);重庆市自然科学基金资助项目(CSTC2010BB221)

Self-adaptive Illumination Invariant Feature Extraction Method for Texture Image

SHANG Zhao-wei 1, HU De-heng 1, ZHAO Heng-jun 1, YANG Jun 1,2   

  1. (1. College of Computer Science, University of Chongqing, Chongqing 400030, China; 2. Linzhi Military Subarea of PLA, Linzhi 860100, China)
  • Received:2012-07-16 Online:2013-11-15 Published:2013-11-13

摘要: 为降低光照变化对纹理图像的影响,提出自适应的纹理光照不变特征提取方法。利用小波变换提取对数域纹理图像的高、低频分量,并分别采用不同方法对两者进行处理,提取其光照不变量图像,运用主分量分析法得到光照不变量数据的特征,使用K-最近特征线分类器进行图像分类。实验结果表明,该方法在光照条件复杂的Outex 14数据集上能够取得较好的分类效果,分类正确率高于现有方法5.56%~22.10%。

关键词: BayesShrink算法, 小波去噪模型, 自适应, 主成分分析, K-最近特征线分类器, 光照不变特征

Abstract: To reduce the influences imposed by changing illumination on texture images, a self-adaptive scheme is proposed to extract illumination invariant feature for texture images. Wavelet transform is utilized to extract both high and low components of the logarithmic images. Different processing strategies are each applied to each component to gain the illumination invariant image. Primary component analysis is adopted to obtain the illumination invariant feature. And a K-nearest feature line classifier is employed for classifying. Experimental results show that on Outex 14 texture dataset, the performance of the method is better than that of existing methods, from 5.56% to 22.10% higher respectively.

Key words: BayesShrink algorithm, wavelet denoising model, self-adaptive, principle component analysis, K-nearest feature line classifier, illumination invariant feature

中图分类号: