AN ENSEMBLE OF OPTIMAL DEEP LEARNING ARCHITECTURE WITH RANDOM FOREST CLASSIFIER FOR CONTENT BASED IMAGE RETRIEVAL SYSTEM

Content based image retrieval (CBIR) extract the details from the images depending upon the content exist in the image as feature descriptors. It intends to the process of retrieving images with maximum resemblance among the visual content exist in the massive databases. Feature extraction and similarity measurement are the two essential steps involved in CBIR. This paper presents an ensemble of optimal AlexNet architecture with random forest (RF) classifier called EOANA-RF model to effective retrieve the images from the databases. Since AlexNet model does not offer superior results on large databases, the AlexNet architecture is optimized in three ways. Firstly, the average pooling undergoes replacement with max-a ve pooling, Maxout is employed in fully connected (FC) layers and hidden layer is included to map high-dimension features to binary codes. Then, similarity measurement and RF based classification process takes place to retrieve the images related to the query image (QI) from the databases and classifies it. The performance of the proposed model undergoes validation using CorellOK dataset. The obtained simulation outcome verified the enhanced performance of the proposed model on the applied dataset.