Classification of images using Extreme Learning Machine and Regularization Extreme learning Machine (ELM) with the leave-one-out cross validation

Image classification is an essential task, which can be considered as a central point to handle image datasets for improving the accuracy of image retrieval. Traditional image classification methods commonly utilize Support Vector Machines (SVM) as image classifiers. On the other hand, there are several disadvantages of using SVM, such as the high computational cost and large number of parameters to be optimized. In this paper we propose an Extreme Learning Machine (ELM) based regularized extreme learning machine with leave-one-out (RELM-LOO) to improve the accuracy of classification using Mammographic Image Analysis Society (MIAS) mammography breast cancer images and USPS digits datasets. The leave-one-out cross validation LOO approach to optimize the regularization parameter of ELM and accordingly the best classification hyperplane is attained. Experiments on the widely used MIAS breast cancer mammography datasets and USPS digits database images demonstrate that our approach can effectively improve the accuracy of image classification and achieve performance at extremely high speed. There are no works where such modification is applied to mammography breast cancer images. Results obtained in terms of classification accuracy and training time compared to the original ELM and regularized ELM using LOO With USPS digits dataset, it achieved 89.23% in Extreme Learning Machine, 94.18% in Artificial Neural Network ANN-Adam and 94.47% in regularized ELM via leave-one-out (RELM-LOO). Further improved accuracy and reduced computational complexity.