基于不变矩和SVM分类的三维目标识别方法

Multi-view 3-D Target Recognition Algorithm Based on Moment Invariant and Support Vector Machine Classification

在计算机视觉问题的研究中,针对三维目标识别,可综合应用图像的不变矩特征和支持向量机分类方法,为快速目标识别,减少计算量,提出了一种红外图像中多视点目标的识别方法。首先获取各类三维目标的若干二维视图,将视图放在一起进行标准化处理并提取它们的不变特征矩。然后对每组视图的Zernike矩进行聚类;将聚类中心对应的Zernike矩作为此类飞机的特征矩,就完成了三维目特性视图的选取。识别过程中,针对实际要识别的目标,提取它的特征矩并应用支持向量机的方法进行多目标分类。测试结果表明,提出的方法较好地实现了红外图像中多角度目标的识别准确性,与传统的三维目标识别算法相比,计算量较小,是一种有效的自动目标识别算法。

Synthetically utilizing image invariant moments feature and SVM （Support Vector Machine）classifica- tion method, a novel recognition algorithm was proposed to deal with multi-view target in infrared images. Firstly, many 2-D viewers of every kind of target are acquired. Putting all these views together, we normalize every view＇ s transformation and moment. Then clustering the Zemike moment of every kind into several classes. Taking the Zemi- ke moment responding the center of these classes as the feature-moment of certain kind of plane, we have extracted the 3-D Target feature-views. When real-time target arrives,its features are extracted and pair-wise SVM classifier was used to realize the multi-target classification. A large number of recognition tests on multi-view targets in infrared images prove the validity and reliability of the scheme in this paper.