Multi-level convolutional channel features for content-based image retrieval

Kun Hu, Yuan Dong, Hongliang Bai

Abstract: An effective content-based image retrieval (CBIR) system depends on the discriminative feature which represents an image. In this work, we explore deep convolutional features for a CBIR system. We first show the effectiveness of deep convolutional channel features for a CBIR system. Then we introduce a Multi-Level Pooling method (MLP) to obtain object-aware features from convolutional layers and finally the features extracted from different layers are incorporated to a short representation vector. Through multiple experiments, we show that our approach can achieve state-of-art results on several benchmark retrieval datasets.

External IDs:dblp:conf/vcip/HuDB16