Evaluation of Different Features and Classifiers for Classification of Rays from Underwater Digital Images

Machine learning technique based classification approaches have proved credibility and gained popularity in recent years. Different feature extraction methods from twodimensional images along with machine learning classifiers are being used for different computer vision problemsolving approaches. To address the challenges of image classification, sourced from underwater imaging, we have experimented on some prominent feature extraction algorithms namely histogram of oriented gradients (HOG) , local binary patterns (LBP), bag-of-features (SURF), and neural network based feature extraction on a dataset containing total 305 underwater images of blue spotted and manta rays. As classifiers, we have experimented on decision tree, support vector machine (SVM), k-nearest neighbours (KNN), and neural network (two-layered NN). Results show that the combination of bag-of-features with twolayered neural network classifiers performed better than others for the selected underwater image dataset. As a tool, we used Matlab for this experimentation.