Fusion of entropy-based color space selection and statistical color features for ripeness classification of guavas

Abstract

This paper presents a novel and non-destructive approach to the color appearance characterization and classification of guava ripeness. Guava ripeness is modeled using extracted statistical color features and support vector machines (SVM) are adopted to perform the classification task. Also, the role of different color spaces in entropy calculation for estimating resolving power in the characterization of ripeness levels of guava is investigated. This approach is applied to 270 guava images from three types of ripeness, i.e., under ripe, ripe, and over ripe. Entropy-based color space selection is carried out using nonparametric Kruskal– Wallis procedure. Statistical curve-fitting color features are derived from the histogram of selected color space. Experimental results show that in spite of the complexity and high variability in color appearance of guava, the modeling of guava images with statistical color curve-fitting parameters allows the capture of differentiating color features between the guava ripeness levels. The classification accuracy using six normpdf curve-fitting parameters (mean, sigma, mean_LB, mean_UB, sigma_LB, sigma_UB) is 90.37% for testing data.