Analysis and design of neural network based image retrieval system for identification of species

Abstract

Due to the advancement in image acquisition and data storage methods, the size of image database has been increasing. Manual annotation of images and textual description was earlier used for image retrieval which was time consuming. The need of the hour is to manage the large collections through efficient systems called content based image retrieval systems. Instead of the data associated with the image, the visual contents of the image are analysed such as colour, arrangement and shape of the objects present in the images. These systems are more efficient and fast as compared to the conventional method of image retrieval. In this paper, a neural network based image retrieval of species has been explained. The segmentation techniques such as morphological operations have been performed on the images and feature extraction techniques have been implemented for the retrieval of images. The neural networks are being used for the solution of a wide variety of computer vision problems. In this work, the features extracted are fed to the neural network which is then trained. The feature vectors of the query image and database images are then compared. The execution of the method proposed above has been carried out in MATLAB and the images retrieved are indexed on the basis of similarity measurement between them. The system developed has shown a remarkable improvement in the retrieval results as compared to the existing systems.