Statistical features extraction of discrete curvelet transform for surface quality evaluation of mangosteen

Fast discrete curvelet transform can be used in differentiating between good and defect of mangosteen surfaces. However, the transformed surface image need to be extracted by features extraction methods to be used by Linear Discriminant Analysis (LDA) for detecting whether the surface condition is a defect or not. In this paper, we test commonly used extraction methods consist of mean, energy, entropy, standard deviation, variance, sum, correlation, contrast, and homogeneity to see which are suitable to be used in detecting mangosteen surface defects. Furthermore, we use K-Fold Cross Validation method to check the accuracy and 120 images as test materials. Finally, the highest accuracy is shown standard deviation by 91,7% and followed by the variance by 88,4%.