摘要
目前大多数基于形状的物体识别模型中,最大难题就是如何有效应对物体尺度的变化。对此提出一种图像的方向检测算法。该算法能够从复杂图像提取出满意的方向图。与传统算法得到的边缘图、直线图等相比,方向图突出显著语义特征而抑制琐碎干扰信息,更接近物体的真实轮廓图。在此基础上,结合物体的"朝向特征"和"形状特征",设计了一个基于形状匹配的物体识别模型。其中形状特征利用线段之间的角度信息来保证物体尺度的稳定性,朝向特征是通过视觉通路中的简单细胞和复杂细胞的感受野获得,它与自适应的上下文信息整合在一起,适用于具有复杂背景信息的物体的形状匹配。实验结果表明,在ETZH图库测试的基础上,该模型能够有效提高识别的效率。
At present,the biggest challenge is how to deal with changes of object scale effectively in most of the shape-based object recognition model.In this paper,we put forward an algorithm of detecting orientations in digital images,which can extract satisfactory orientation maps from complex images.Compared with the edge/line maps obtained by traditional algorithms,the orientation maps highlight salient semantic features,while suppressing trivial distractions.Moreover,they are closer to the real contour maps of objects.With the proposed model of orientation computation,we design an object recognition model based on shape matching,which combines the“orientation feature"and“shape feature".The shape feature utilizes the angle information between the line segments to ensure the stability of the object scale.The orientation feature is obtained through the receptive field of simple cells and complex cells in visual pathway,which are integrated with adaptive contextual information for the shape matching of objects with complex background information.The experiments show that the model proposed can improve the efficiency of recognition on the basis of ETZH image library.
作者
樊一娜
郎波
FAN Yi-na;LANG Bo(School of Information Technology,Beijing Normal University,Zhuhai,Zhuhai 519087,China)
出处
《计算机技术与发展》
2018年第4期82-86,共5页
Computer Technology and Development
基金
国家自然科学基金(61272364
61375122)
广东高校重大项目与成果培育计划(2016GXJK192)
关键词
朝向特征
形状特征
上下文信息
物体识别
orientation feature
shape feature
contextual information
object recognition