A Comparative Analysis of Multiple Label Image Classification Techniques

Abstract

Multiple label image classification deals with the problem where each image contains a single instance but it is associated with multiple number of object labels. Multiple object label classification problem has attracted significant amount of interest from many people and successfully applied in many image object labelling and object detection, prediction tasks. Since the past ten years, significant number of research development has been done in this machine learning domain. Several conventional neural network approaches applied on multiple object label learning and prediction based on dependency between label of the image object instance, co-existing labels and ranking of multiple labels in the image. This study aims to provide a detailed review analysis on multiple labels learning domain with emphasis on statistical machine learning, object detection and semi supervised multiple label learning using conventional machine learning algorithms and deep learning algorithms. At first, the basics of multiple object label classification such as definition and metrics for evaluation are discussed. Second, many multiple label image classification methods are shortlisted based on common term along with their performance analysis and discussions. Third, several related approaches for classification and setups are discussed and finally open research problems identified and few proposed solutions in the field of multiple label image learning are given for future references.