Figure spotting in Indian heritage image

Figure spotting is one of the important applications in the field of content-based image retrieval. With the recent advances in 3D shape analysis, Wave Kernel Signature (WKS), a kernel-based feature descriptor under the foundation of quantum mechanics performs well than the other kernel based feature descriptors. In this paper, we adopt the WKS as a 2D local patch descriptor for figure spotting. An effective search technique is developed to spot the regions of interest within an image for a given query image. We also use the classical feature descriptors such as scale-invariant feature transform (SIFT), speeded up robust features (SURF), and the histogram of oriented gradients (HOG) for figure spotting and compare their performances. The proposed technique is tested on a dataset which contains 594 images collected from two heritage temples. The performance of the proposed technique is measured using standard evaluation metrics and shows promising results of the proposed method.