基于SVM与颜色/纹理组合特征的景物识别算法  

Natural Object Recognition Algorithm Based on SVM and Color/Texture Combination Features

作　　者：雷宝权[1] 杨丽华[1] 程咏梅[1] 赵春晖[1] 吴燕茹[1] 

机构地区：[1]西北工业大学自动化学院,西安710072

出　　处：《计算机科学》2009年第10期274-276,283,共4页Computer Science

基　　金：国家自然科学基金重点资助项目(60634030);高等学校博士学科点专项科研基金(20060699032);航空科学基金(2007ZC53037)资助

摘　　要：受多种因素影响,室外场景变换复杂,因此利用单个特征(通常多使用颜色或纹理)完成室外场景的识别,不能达到满意的效果。首先基于RGB空间的颜色直方图进行颜色特征提取,然后基于Gabor滤波器进行纹理特征提取,最后将两种特征结合,提出了基于SVM与颜色/纹理组合特征的景物识别算法。基于美国加州理工学院的Pasadena Houses2000数据库建立了室外场景中天空、道路、房子、树木和草地等5类样本训练库,进一步完成了基于SVM的单一特征和颜色/纹理组合特征的景物识别。实验结果表明,该算法对仅用一种视觉特征无法区分景物的室外场景图像能取得较好的分类结果。Affected by many factors,outdoor scenes vary greatly,so using a single feature（color or texture） to complete the outdoor scenes recognition can not achieve satisfactory results. A natural object recognition algorithm based on SVM and color/texture combination features was presented. Firstly, the color histogram based on the RGB color space was extracted as color feature. Then, the texture feature was extracted based on multi-channel Gabor filters. At last, the color/texture combination features were presented. An image database of training samples including sky, road, house, tree and grass was created, which is obtained from Pasadena Houses2000 database of California Institute of Technology. And the natural object recognition based on SVM using a single feature and color/texture combination features was completed respectively. Experimental results show that this algorithm has good recognition effect on the images in which each natural object varies greatly from each other.

关 键 词：景物识别 GABOR滤波器 颜色直方图 颜色/纹理组合特征 SVM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]