Evaluation of Mangosteen Surface Quality using Discrete Curvelet Transform

Abstract

Mangosteen is the major commodity of fruit export from Indonesia. Only free-defect mangosteens are exported which were conventionally classified by human vision. In order to automate the classification between defect and free-defect mangosteen surface and handle high volume of export, machine vision has a great opportunity. The objective of this paper is to classify mangosteen surface images using discrete curvelet transform (DCT). The curvelet transform is a multiscale directional transform, which allows an optimal non-adaptive sparse representation of objects with edges. The methodology of this research involved pre-processing, implementation of DCT, statistical features extraction and classification using linear discriminant analysis. The method has been implemented on a number of 80 mangosteen images and validated using 4-fold cross validation method. The highest accuracy of classification between defect and non-defect surface is 92.5% obtained on second scale of DCT. In conclusion, the proposed method is able to evaluate mangosteen quality surfaces.