Sketch Based Image Retrieval

With the rapid growth of the online multimedia information,how to make the content based image retrieval(CBIR)quickly and efficiently from the massive online datasets becomes a hotspot in the multimedia technology.CBIR is a common application,but it generally requires a limited input image query,and it is usually highly sensitive to the light condition and other factors with a low robustness.With the popularity of touch-screen devices,such as smart phone,pad and so on,Sketch Based Image Retrieval(SBIR)has attracted more and more attention.Considering the limitation of the traditional descriptors such as HOG and SIFT,we propose a novel feature descriptors based on the bag of mid maps of the Convolutional Neural Network(CNN).The main contents of this paper are as follows:1? In this paper,we first make a lot of research on the differences between sketch and the general image,then review the theory of sketch based image retrieval technology in details,and analysis of their respective advantages and disadvantages.2? In this paper,we extract the boundary probability images of the general image and convert the boundary probability images into 3 levels in order to release the domain gap between sketch and the general image.3? We make use of the Chamfer distance transform to add the distance information into the input query sketch and the binary edge map.4? We proposed a novel image descriptors called the bag of mid maps descriptor for final retrieval.We evaluate our proposed descriptor and retrieval strategy on the Flickr15 K datasets.Results show that the proposal achieves significant improvements over the state-of-the-art approaches.