Learning Food Image Similarity for Food Image Retrieval

被引：16

作者：

Shimoda, Wataru ^{[1
]}

Yanai, Keiji ^{[1
]}

机构：

[1] Univ Electrocommun, Dept Informat, 1-5-1 Chofugaoka, Chofu, Tokyo 1828585, Japan

来源：

2017 IEEE THIRD INTERNATIONAL CONFERENCE ON MULTIMEDIA BIG DATA (BIGMM 2017) | 2017年

关键词：

deep learning; Siamese network; triplet network; food image recognition;

D O I：

10.1109/BigMM.2017.73

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

For food application, recipe retrieval is an important task. However, many of them rely on only text query. Food image retrieval has relation to recipe retrieval so that similar food images are expected that they have similar recipes. Rising image retrieval performance is desired for recipe retrieval. On the other hand, to learn similarity by Siamese Network or Triplet Network are known as an effective method for image retrieval. However, there are no research for food image retrieval using similarity learning with Convolutional Neural Network as far as we know. Food recognition is known as one of fine-grained recognition tasks. Therefore it is unclear that how effective similarity learning methods based on CNN in food images. In our work, we trained some networks for feature similarity, and evaluated their effectiveness in food image retrieval.

引用

页码：165 / 168

页数：4