Local and Global Feature Fusion Based Visual Concept Detection in Images

With the advent of digital cameras and mobile phones, advances in telecommunication and internet, millions of images are uploaded on the internet without much information about the image. An efficient method is necessary for automatic image annotation and indexing for the vast collection of images. Concept detection is task of detecting concepts present in image. In this paper, concept detection is obtained by effectively fusing local feature descriptors and global features descriptors. First object extraction is carried out using edge and color, and the aspect ratio of each extracted object is calculated. The local features of all extracted objects and global features of the image are computed. The detected concept of the query image is displayed based on the local and global feature matching scores obtained using our algorithm. The proposed algorithm is evaluated on Wang’s Corel dataset consisting of 1000 images. Results demonstrate that the proposed approach outperforms the KNN and ANN methods with high accuracy.