A Review on Label Prediction through Multiple Visual Features

Abstract

Multiple visual features are represented by multimedia data. Multi-feature learning aims at using the complementary structural information of visual features. The focus is on the semi-supervised learning when the label information of the training data is insufficient. Most of the existing systems face the problem of insufficient labelled data that are expensive to label by hand in real-world application. To address this problem classifier has been already proposed in the literature that select features closely similar to the query image and based on these features label prediction is done. This work aims at studying different low-level feature descriptor for better feature extraction and focusing on computational time of the system by replacing SIFT descriptor by ORB feature descriptor