An Improvement Method of Kd-Tree Using K-Means and K-Nn For Semantic-Based Image Retrieval System

In this paper, an improved method of KD-Tree using k-Means and k-NN is proposed and applied to a semantic-based image retrieval system. Firstly, the image data is clustered by the k-Means algorithm at each layer on KD-Tree. Secondly, a process of image classification according to the k-NN algorithm is conducted at leaf nodes. So, an algorithm of KD-Tree construction based on the k-Means and k-NN is implemented. Each query image is extracted to a feature vector and performed image classification on KD-Tree. Then, a SPARQL query is created to retrieve a set of similar images by semantics on ontology. Based on this theory, a model of semantic-based image retrieval is proposed. The experiment is performed on Wang, Caltech101, and Caltech256 data sets. The experiment results are compared with the published works on the same data set to demonstrate the efficiency of our proposed method.