Abstract: In order to enhance the robustness on affine transformation and reduce false recognition rate of object recognition,an object recognition method based on Affine Scale-Invariant Feature Transform(ASIFT) and graph matching is proposed.It detects key points in the image by using ASIFT method,extracts the local features of each key point,including ASIFT features,texture features and color features,and uses three-tuple graph to represent these features.Objects are recognized by using separate layer matching,which executes graph matching after successful key point matching to reduce the false recognition.Parameter selection and recognition performance of the new method are evaluated comprehensively on two common datasets including PVOC-2007 and COIL-100.Results show that the new method has higher recognition rate and is a valid object recognition method.

Key words: object recognition, affine transformation, Scale-Invariant Feature Transform(SIFT), Affine SIFT(ASIFT), graph matching

摘要： 为提高物体识别对仿射变换的鲁棒性并降低误识别率,提出一种新的物体识别方法。采用仿射-尺度不变特征变换(ASIFT)方法检测图像中的关键点,提取各关键点的局部特征,包含ASIFT特征、纹理特征和颜色特征,并用三元图表示融合后的特征。采用分层匹配的思路识别物体并进行关键点匹配,匹配成功后再进行图匹配以降低误识别率。在PVOC-2007和COIL-100 2个公共测试数据集下对该方法的参数选取和识别性能进行综合评价,结果表明,该方法具有较高的识别率,可有效进行物体识别。

关键词: 物体识别, 仿射变换, 尺度不变特征变换, 仿射-尺度不变特征变换, 图匹配

CLC Number: 

• TP391