摘要
为解决当前服装款式识别领域中,服装轮廓特征提取技术较复杂,其分类方法的效率低、适应性差等问题,提出一种新型的服装款式的识别方法。首先创建了一个服装图像样本库,并从这些服装图像中提取服装轮廓,然后使用傅里叶描述子描述服装的轮廓特征,以多分类支持向量机进行分类。结果表明,该方法能够准确提取服装轮廓,傅里叶描述子的识别效果优于Hu不变矩和融合特征(Hu不变矩和傅里叶描述子);对傅里叶描述子进行主成分分析不能提高识别准确率;支持向量机的分类效果优于极端学习机;该方法能够达到95%以上的识别率,尤其对轮廓特征明显的款式有更好的识别率。
In the current clothing style recognition field, clothing contour feature extraction technique was complicated, classification efficiency was low and adaptability was poor. In order to solve these problem and recognize the clothing styles, a novel approach was proposed. In this approach, the contours were extracted from the clothing images, which were taken from the newly created sample database. Then the contour features were described by Fourier descriptors(FD). Finally, the clothing styles were classified by multiclass support vector machines(SVM). The experimental results show that this novel approach can accurately extract the contours of clothing. The recognition effect of the Fourier descriptors is better than the Hu moment invariant and feature fusion (Hu moment invariant and Fourier descriptor). Principal component analysis of FD can’t improve the recognition accuracy, and the classification effect of SVM is better than ELM. This approach can achieve a recognition rate above 95%. In particular, contour features obvious style has a better recognition rate.
出处
《纺织学报》
EI
CAS
CSCD
北大核心
2017年第5期122-127,共6页
Journal of Textile Research
关键词
服装款式识别
傅里叶描述子
支持向量机
HU不变矩
主成分分析
极端学习机
clothing style recognition
Fourier descriptor
support vector machine
Hu momentinvariant
principal component analysis
extreme learning machine