Fusion of Contourlet Transform and Zernike Moments Feature Extraction using DNN and ELM Classifiers for MRI Brain Tumor Images
Abstract—Abstract Image retrieval is that the most vital application that has been used extensively in image processing. Content Based Image Retrieval (CBIR) is employed to search and retrieve the expected image from the database. Magnetic resonance imaging (MRI) technique plays a crucial role in diagnosing many diseases in human brain. In this paper, we propose a fusion technique for T1 and T2 weighted MRI scans. Our proposed technique has three parts. First, texture and shape features are extracted from a brain tumor images. Next, the fusion techniques like genetic algorithm (GA) and particle swarm optimization (PSO) are used to combine the texture and shape features. Finally, the popular supervised learning machine techniques like Deep neural network (DNN) and Extreme learning machine (ELM) are used to classify the brain tumor based on the selected features. The experiment is done on 1000 brain tumor images. Six measures, namely sensitivity, specificity, accuracy, error rate, Jaccard coefficient and f-measure are used to evaluate the performance of the method. Experiment results show that the average sensitivity for DNN – 51.48%, ELM- 51.33%, specificity for DNN- 40.19%, ELM- 47.6%, error rate for DNN-16.12%, ELM- 4%, Jaccard coefficient for DNN- 49.72%, ELM- 50.48%, f-measure for DNN- 65.09, ELM- 67%, accuracy for DNN – 88.8% and ELM – 96%.
sales on Site11,021