Image Based Indonesian Fruit Recognition using MPEG-7 Color Structure Descriptor and k-Nearest Neighbor
| Main Authors: | Siswantoro, Joko, Arwoko, Heru, Widiasri, Monica |
|---|---|
| Format: | Proceeding PeerReviewed application/pdf |
| Terbitan: |
, 2019
|
| Subjects: | |
| Online Access: |
http://repository.ubaya.ac.id/36445/1/Paper%20InCITE%202019%20Joko.pdf http://repository.ubaya.ac.id/36445/ |
Daftar Isi:
- Image based fruit recognition can be applied in several sectors including food industry, food retail, and medical. This paper proposes a method to recognize Indonesian fruit from image. The method uses MPEG-7 Color Structure Descriptor (CSD) as input features to k-nearest neighbor classifier. CSD describes the local color structure of image in HMMD (Hue, Max, Min, and Difference) color space. In this study, the numbers of features extracted from a fruit image were 32, 64, 128, and 256. A simple feature selection method based on variance has been applied to reduce the dimension of input features and to increase classification performance. A feature with variance less than predefined threshold was excluded from feature space. Three hundred and fifty images from seven types of Indonesian fruit have been used to validate the proposed method using 10-fold cross validation. The experimental result showed that the best classification accuracy of 90.86% was achieved using 256 features of CSD combined with feature selection.