The Effect of Block Size, Training Set and K-Value in the Classification of Food Grains Using HSI Color Model

The method proposed for the classification of food grains can be divided into two phases: 1) color, global and local feature extraction 2) classification using extracted features. This paper presents the effect of block size, training set and K-value in the classification of food grains using HSI color model by combining color and texture information without pre-processing. The first phase in the proposed system is feature extraction. The features are computed locally and globally. The co-occurrence matrix helps to extract features locally. The non-uniformity of RGB color space is eliminated by Hue, Saturation and Intensity (HSI) color space. Further, minimum distance and K nearest neighbour algorithms are used for classification. Percentage of correct classification and error analysis are carried out by confusion matrix. The proposed work attains maximum average accuracy of 95.83% and 70.37% for block size 512×512 and 256×256 with K=5 and K=1 respectively.