Content-Based Image Retrieval Using a Fused Descriptor

Content-based image retrieval (CBIR) refers to the techniques that are used for retrieving similar image from a given set of images for a particular image. The paper proposes a unified framework for CBIR that relies on texture, shape and global descriptors for image retrieval. In order to retrieve similar images, a variety of descriptors such as histogram of oriented gradients, local binary pattern and GIST are calculated and combined together. These descriptors are calculated separately for the training and test databases. Further, various distance metrics are used for computing the closeness of the input images with the retrieved images. The proposed method, when tested on the Wang image database comprising 1,000 images, outperforms other state of the art techniques.