In the recent work proposed Audio Video Retrieval algorithm (AVRA) is necessary for efficient search of media using Multiplication Kernel Support Vector Machine (MUK-SVM) classifier. Though it produces higher retrieval results, MUK-SVM algorithm includes many vital parameters, which have to be rightly fixed to attain the remarkable classification outcomes for AVIAE. Which consists of consists of four major steps. Initial step is accountable for automatic video structuring. It significantly spotlights on extracting Texture, Video Key Frame Indexing, audio features, Color, etc utilized for AVISA. Subsequent step store structure as video creating Video Table of Contents (VToC) in eXtensibleMarkup Language (XML) format, and offer VToC on Web with eXtensibleStylesheet Language (XSL). Final one offers user’s options to personalize video into summary utilizing Fuzzy C Means with Gaussian Kernel Distance (FCM-GKD) based clustering, then the summarization of video is to be stored in the database is known as indexing it also gets from the feature extraction terms. Finally the query result from the user for the retrieval using the classification algorithm called Hybrid Deep Learning, where the artificial bee colony (ABC) technique is introduced in the form of an alternate mechanism for the performance optimization of a Convolutional Neural Network (CNN). Otherwise said, the proposed method is aimed at reducing the errors in classification through the initialization of the CNN classifier weights depending on the solutions that the ABC technique yields. Also, the distributed ABC is also suggested as a technique for maintaining the amount of time required for the process execution while dealing with massive training datasets. AVRA framework is developed using MATrixLABoratory (MATLAB) and measured in terms of precision, recall, f-measure and accuracy.
Volume 12 | 01-Special Issue
Pages: 641-653
DOI: 10.5373/JARDCS/V12SP1/20201113