Clustering based Indexing of Celebrity Cartoon Images for Retrieval

Celebrity cartoon image has its own attention as it characterizes a famous personality and also serves as a good platform to communicate the scenario, express emotions and makes it easy to perceive without much effort. Retrieval of celebrity cartoon images based on a query cartoon image is a difficult task because of the presence of caricatures, sketches, drawings, painting and artistic style variations. Matching cartoon images and retrieving them is a challenging task as artists may vary the same image with different styles and shape exaggerations for the same feature. A novel clustering based tree is constructed with different levels and each level with varied number of nodes. The nodes are expected to have similar images which are identified using partitional clustering. Tree based indexing is created by designating unique indices for each node. The indices are framed in such a way that they follow their parent node. The given query image traverses the tree till leaf node to retrieve the images based on a distance threshold. If the required number of images are not retrieved, then backtracking is done to retrieve the similar images in the other children nodes of same parent which are reached from the root based on the distance. The IIIT-Cartoon Faces in the Wild (IIIT-CFW) dataset is taken for the experimentation. Images considered in this dataset contains caricatures, paintings, cartoons and sketches of celebrities. Through analysis, it is found that the proposed approach performs well in terms of evaluation measures as precision and recall, time and number of relevant retrieved images. Also it can be used in reduction of search space in recognition of images as the retrieved relevant images could be used as a competent search space. Hence, the proposed approach aids subsequently the celebrity cartoon image recognition. The proposed approach is compared against two models as retrieval using linear search and linear search based on class representative.