Optimization of Multi-Class Non-Linear SVM Image Classifier Using A Sobel Operator Based Feature Map and PCA

Humans are skilled at categorizing things fast. Automation of this skill becomes beneficial for various applications. SVM is a useful machine-learning algorithm for image classification. But sometimes, training large datasets on SVM classifier can be time-consuming and computationally extensive, while not giving good accuracy. In this study, we have used dimensionality reduction using PCA for a multi-class SVM image classification model which uses a custom feature map based on Sobel operator. The kernel used in the SVM model is non-linear type. This makes the procedure highly efficient by using the mentioned feature extraction method and the dimensionality reduction procedure.