基于迁移学习与支持向量机的服装舒适度评估

针对传统服装舒适度评估需要直接试穿服装导致的舒适度评估精确度不高和评估过程耗时的问题,提出一种从试穿服装数据库学习服装舒适度评估模型的方法,可以快速准确地评估服装舒适度。首先,采集试衣模特尺寸和试穿样板图,并利用迁移学习改善试穿样板图构建试穿服装数据库,同时提出基于虚拟试衣技术的舒适度标签获取方法,为数据库中对应的试穿样板图添加舒适度标签;然后,提取试穿样板图的局部二值模式为服装样板特征,并融合试衣模特尺寸数据形成服装试穿特征向量;最后,提取试穿服装数据库的融合特征训练支持向量机,得到服装舒适度评估模型。实验结果表明,该方法的准确率和系统时间分别为0.8344和12 s,具有较高的精确度和效率。

Clothing comfort evaluation based on transfer learning and support vector machine

The traditional methods for clothing comfort evaluation is carried out through the try-on effect of the garment, which requires much time but with low evaluation accuracy. This paper presented a clothing comfort evaluation model learning from clothing patterns based on the transfer learning and support vector machine fast and accurately. The sizes of mannequins and the graphs of garment patterns were firstly collected, and the graphs of garment patterns were improved by using transfer learning to create garment pattern database. Then, a comfort label acquisition method was presented based on Virtual Try-On, adding comfort label to the corresponding graph of garment pattern. Following that, local binary pattern was extracted from the graph of garment pattern, and it was combined with the sizes of the corresponding mannequins to form clothing comfort feature vector. Finally, the clothing comfort feature vectors of garment pattern database were extracted to train the support vector machine. This exercise shows that the accuracy and average time to evaluate clothing comfort using this method are 0.834 and 12 s respectively, representing satisfactory accuracy and efficiency.