深度网络特征在视频走秀场景下的匹配应用
首发时间:2016-11-01
摘要:互联网和电子商务的高速发展给计算机视觉带来了新的应用场景,服装是电商平台交易中的第一大类目。本文将深度学习特征提取应用到视频走秀场景下的服装匹配,分别提取浅层图像特征与深度网络特征进行对比分析,浅层特征中对hog特征做出了扩展,提出了hog+颜色直方图的新方式,进而进行自动化匹配推荐。实验表明,级联的浅层特征与深度网络特征都比单一浅层特征有更好的准确率,且深度网络特征的效果最佳,其在测试视频中平均准确率达到68.8%。?
For information in English, please click here
Clothes Matching Of Online Show Video Based on Deep Learning Feature
Abstract:The swift development of Internet and Electronic Commerce has brought new application scenaries and Clothes is the biggest kind of Electronic Commerce. In this paper, we applied deep learning to Clothes Matching of Online Show video and compared effect of shallow image features and deep learning features. We cascade hog with color histogram to perform colthes matching. Experiments show that traditional cascade feature and deep learning feature are more useful than a single traditional one, and the deep learning feature performs best, which achieves 68.6% average accuracy in the test videos.
Keywords: pattern recognition deep learning clothes matching video show
基金:
论文图表:
引用
No.4707242115623214****
同行评议
共计0人参与
勘误表
深度网络特征在视频走秀场景下的匹配应用
评论
全部评论0/1000