A COMPARISON OF GRAY-LEVEL RUN LENGTH MATRIX AND GRAY-LEVEL CO-OCCURRENCE MATRIX TOWARDS CEREAL GRAIN CLASSIFICATION

This study describes a comparison of texture features based on Gray Level Co-Occurrence Matrix (GLCM) and Gray Level Run Length Matrix (GLRLM) towards bulk grain classification. In order to have a fair comparison, four features were extracted each from GLCM and GLRLM. A total of 400 bulk grain colour image were taken for 4 different verities of rice (100 for each rice type). A Back-Propagation Neural Network (BPNN) with adaptive thresholded output is used for classification. The network was trained with 200 image samples (50 for each rice type) and the same is tested with all the 400 images. It is found that the average classification accuracy base on GLRLM texture features using BPNN is 99.5%, which is better as compare to that of GLCM texture features having a average classification accuracy of 97.75%. The classification accuracy using BPNN was also compared with other classifiers like K-NN and SVM. Results shows that BPNN provides better consistency in classifying rice grain as compare to K-NN and SVM