Ensembling Framework for Texture Extraction Techniques for Classification

In the past few years, texture-based classification problems have proven their significance in many domains, from industrial inspection to health-related applications. New techniques and CNN-based architectures have been developed in recent years to solve texture-based classification problems. The limitation of these approaches is that none of them claims to be the best suited for all types of textures. Each technique has its advantage over a specific texture type. To address this issue, we are proposing a framework that combines existing techniques to extract texture features and displays better results than the present ones. The proposed framework works well on the most of the texture types, and in this framework, new techniques can also be added to achieve better results than existing ones. We are also presenting the SOTA results on FMD and KTH datasets by combining three existing techniques, using the proposed framework.