Improving the Performance of Machine Learning Classifiers for Image Category Identification using Feature Level Fusion of Otsu Segmentation Augmented with Thepade's N-ary Sorted Block Truncation Coding

摘要

Image Classification has become important due to the ever-increasing data in the social industry. Nowadays various Machine learning algorithms are used to classify images based on their category. For the purpose of categorization, various thresholding algorithms are used to extract features from the images. Wang Dataset which consists of 1000 images was used to train the dataset with the help of Machine learning classifiers alias MLP, SMO, Random Forest, Random tree, Bayes Net, Naive Bayes etc. Otsu thresholding algorithm and Thepade's N-ary Sorted Block Truncation Coding (TSBTC) algorithm was used to extract features from the query images. To obtain check for better accuracy, feature level fusion of Otsu and TSBTC was applied on the query images which gave better accuracy as compared to Otsu and TSBTC individually. The trained classifier predicted the category of the query image for which testing options like 10- fold cross validation, 80-percentage split and 50- percentage split were used on Machine learning classifiers to obtain maximum accuracy for image classification.