Multilevel Block Truncation Coding with diverse color spaces for image classification

The paper depicts the use of Multilevel Block Truncation Coding for image classification. Feature vectors are extracted with four levels of Block Truncation Coding to classify the several categories of images for performance comparison in six different color spaces for the proposed methodology. Three databases out of which two are public databases and one is a generic database are considered for the experimentation. The two public datasets used are Coil Dataset and the Ponce Group 3D Photography Dataset respectively. The performance of the proposed classifier is tested on all three databases considered. In each of the considered color spaces improved performance is being observed with increasing levels of BTC and BTC level 4 is proved to be better as compared to other BTC levels. Overall Kekre´s LUV color space has shown the best performance for BTC level 4 based image classification.