基于关键点的服装检索

发布时间：2018-07-12 08:46

本文选题：关键点 + 深度卷积神经网络　；参考：《计算机应用》2017年11期

【摘要】：目前,同款或近似款式服装检索主要分为基于文本和基于内容两类。基于文本算法往往需要海量标注样本,且存在人工主观性带来的标注缺失和标注差异等问题;基于内容算法一般对服装图像的颜色、形状、纹理提取特征,进行相似性度量,但难以应对背景颜色干扰,以及视角、姿态引起的服装形变等问题。针对上述问题,提出一种基于关键点的服装检索方法。利用级联深度卷积神经网络为基础,定位服装关键点,融合关键点区域低层视觉信息以及整幅图像的高层语义信息。对比传统检索方法,所提算法能有效处理视角、姿态引起的服装形变和复杂背景的干扰;同时不需大量样本标定,且对背景、形变鲁棒。在Fashion Landmark数据集和BDAT-Clothes数据集上与常用算法进行对比实验。实验结果表明所提算法能有效提升检索的查准率和查全率。
[Abstract]:At present, the same style or similar style clothing retrieval is divided into two categories: text-based and content-based. Text based algorithms often need a large number of tagged samples, and there are some problems such as missing annotation and annotation differences caused by artificial subjectivity. Based on content algorithm, the color, shape and texture features of clothing images are generally extracted, and the similarity is measured. However, it is difficult to deal with the background color interference, as well as the angle of view, posture caused by the deformation of clothing and so on. In order to solve the above problems, a key point based clothing retrieval method is proposed. Based on cascaded deep convolution neural network, the key points of clothing are located, and the low-level visual information of the key points and the high-level semantic information of the whole image are fused. Compared with the traditional retrieval method, the proposed algorithm can effectively deal with the disturbance of garment deformation and complex background caused by visual angle and posture, and it does not require a large number of samples to calibrate, and is robust to background and deformation. The experiments are carried out on Fashion Landmark dataset and BDAT-clothes dataset with common algorithms. Experimental results show that the proposed algorithm can effectively improve the precision and recall of retrieval.
【作者单位】：江苏省大数据分析技术重点实验室(南京信息工程大学);
【基金】：国家自然科学基金资助项目(61622305,61502238,61532009) 江苏省自然科学基金资助项目(BK20160040)~~
【分类号】：TP183;TP391.41

【相似文献】