Search by Detection: Object-Level Feature for Image Retrieval

In content-based image retrieval (CBIR), images are usually represented by local invariant features or global features. Although great success has been witnessed, there still exists some non-trivial problems with those features. In this paper, we propose a novel image representation for image retrieval. We identify some regions of interest with an advanced general object detector, and the regions are described by features extracted with convolutional neural network (CNN). We evaluate the performance of the proposed representation on two public datasets. The experimental results demonstrate the effectiveness of the proposed method.