The second group is based on deep learning models, par- ticularly deep convolutional neural networks (CNNs), which have been successfully used in various visual tasks such as image classification, object detection and image retrieval. Image representation and similarity measure become critical to image retrieval task which aims to find matched images in a big dataset. The recent works [3], [4] proposed an idea ofutilizing a pretrained CNN from a related image classification problem in image retrieval and showed very promising results.