Ensemble Improved Images Classification Based on Unified ELM

Abstract

In this digitalized world we are always on the verge of mixing up similar documents leading to the discrepancy of the user. Visual Impact to human beings are one of the most permanent and inscribed memories lasting through time. Hence, images needs to be classified to be able to serve the purpose of distinct identification. This project aims to classify the images using unified ELM network. With the use of supervised, semi-supervised and unsupervised learning based on MapReduce framework and Dropout and DropConnect algorithm, the classification results with high computational efficiency, can be reached. Unified ELM uses the weighted-matrix method to initialize random values to fasten the throughput. Thus, through our proposed technique, the locality encoding methods of the algorithms being used can lead to effective approaches of improving the ever challenging large-scale image classification problems.