Retrieval of Videos of Flowers Using Deep Features

This paper presents an algorithmic model for the retrieval of natural flower videos using query by frame mechanism. To overcome the drawback of traditional algorithms, we propose an automated system using a deep convolutional neural network as a feature extractor for the retrieval of videos of flowers. Initially, each flower video is represented by a set of keyframes, then features are extracted from keyframes. For a given query frame, the system extracts deep features and retrieves similar videos from the database using k-nearest neighbor and multiclass support vector machine classifiers. Experiments have been conducted on our own dataset consisting of 1919 videos of flowers belonging to 20 different species of flowers. It can be observed that the proposed system outperforms the traditional flower video retrieval system.