Texture images classification plays an important role in machine vision that can be used to distinguish the surface and objects of an image from each other. Texture classification is a two-phases process consisting of feature extraction and classification. Feature extraction is a very important step in texture classification. Thus, we use improved local quinary pattern (ILQP) as descriptor to detect texture information of the images in feature extraction phase. Besides, in the classification phase, an innovative ensemble learning-based method is proposed which is named mixture of extreme learning machine-based experts with trainable gating network (MEETG). This method takes the advantages of extreme learning machine (ELM) for designing the structure of mixture of experts (ME) to overcome on some drawbacks of ME such as computation complexity and time-consuming learning process. The performance of texture features and MEETG is evaluated by applying five datasets: Brodatz album, ULUC dataset, KTH-TIPS dataset, Outex TC000012 dataset and ALOT dataset. Experimental results indicate that MEETG outperforms the other ensemble learning methods such as Bagging, Boosting and ME and also outperforms single classifiers such as nearest neighbor, decision tree, multi-layer perceptron and ELM on classification accuracy.
展开▼
