Microstructure Image Classification: A Classifier Combination Approach Using Fuzzy Integral Measure

In recent times, machine learning-based methods have gained popularity in various materials science applications including microstructure image classification. This paper explores the use of classifier combination approaches for classifying the microstructure images with an improved accuracy. Classifier combination methods have been recognized as a state-of-the-art approach to enhance the performance of many challenging image classification tasks. Ensemble methods are used to increase the predictive performance of a learning system by combining the predictive performances of several base learners. In our proposed model, the features of three-class microstructural images are extracted using the rotational local tetra pattern feature descriptor. These features are separately fed to three different classifiers, namely support vector machine, random forest, and K nearest neighbor. Then, a classifier combination approach based on the confidence scores provided by these classifiers using fuzzy measures and fuzzy integrals is applied for the image recognition purpose. Unlike other straightforward classical classifier combination methods, this method nonlinearly aggregates the objective evidences in terms of a fuzzy membership function, with the subjective assessments of the relative importance of different classifiers. The proposed method has also been compared with many standard classifier combination approaches commonly found in the literature. The experimental results support the effectiveness of fuzzy combination to produce higher classification accuracy than that of the best base classifiers and some popular classifier combination methods.