A Cognitive Novel Autoencoder Deep Convolutional Neural Network And Machine Learning Approachesfor Content Based Image Retrieval System

Abstract

Content based image retrieval is a biometric system for recognizing and classifying or retrieving images on the basis of different patterns from a huge content based image retrieval database. In this work, content based image retrieval recognition using transform techniques combined with machine learning algorithms has been implemented. It is generally used in security, medicine, law enforcement, etc. Content based image retrieval recognition usually involves pre-processing, content based image retrieval representation, and distance measurement or classifier. Content based image retrieval system can be implemented in three ways: the transform based methods, the machine learning algorithms, and the deep learning algorithms. Currently, explicit programming is needed for these methods, and there is a demand for prediction methods. To address these problems, machine learning algorithms are implemented such as supervised and unsupervised learning. In this approach, the auto encoder convolutional neural network and pooling are applied to extract hidden features effectively. Computational complexity is less with machine learning algorithms but it fails to address a huge amount of data. A deep learning algorithm is effective with the huge amount of data. The convolutional layer generates a model on the basis of different filters such as edge detection, sharpening, edge enhancement, and blurring. The pooling layers reduce the dimensionality after using the Rectifier Linear Unit (ReLu) activation function. The proposed methods exhibited better performance than the existing methods of the standard databases ORL, Yale, and the local database, in terms of the content based image retrieval recognition performance measures and computational time.

Information

Applicants

Inventors

Specification

Documents

Orders