Markov Random Field Modeling for Image Classification ---Final project report

This project focuses on the Markov Random Field modeling for image classification problem. For most 2D images with reasonable resolutions, pixels have spatial constraints, which should be enforced during the classification. For the sake of computational simplicity, the identical independent distributed (I.I.D.) assumption is commonly used. Due to this assumption, some unreasonable holes will appear. The goal of this project is to use the undirected graphical model (Markov Random Field technique) to improve image classification results without sacrificing too much computation time. With the MRF-MAP framework, experimental results show that better system performance can be achieved.