Exploring improvements for simple image classification

In this study, we explore potential improvements to a simple image classification methodology. The methodology analyzed uses K-means clustering and a linear support vector machine (SVM) to classify images using raw pixel data from the individual images. We explore improvement of classification accuracy through feature augmentation, use of alternatives to K-means clustering, and modifications to the SVM. In particular, we analyze potential classification accuracy improvements of feature augmentation via the addition of data generated by feature detection algorithms, the use of clustering methods which automatically calculate the number of centroids, and the effect of different types of kernels on the over classification accuracy. In our experiments, classification of the CIFAR-10 dataset is used to measure performance changes.