ASALTAG : Automatic Image Annotation Through Salient Object Detection and Improved k-Nearest Neighbor Feature Matching

Theresia Hendrawati (Computer Science Graduate Program Ganesha University of Education (UNDIKSHA) Singaraja, Bali, Indonesia)
Duman Care Khrisne (Electrical Engineering Department, Faculty of Engineering Udayana University (UNUD) Badung, Bali, Indonesia)

Journal of Electrical, Electronics and Informatics
Vol 2 No 1 (2018): JEEI (February 2018)

Image databases are becoming very large nowadays, and there is an increasing need for automatic image annotation, for assiting on finding the desired specific image. In this paper, we present a new approach of automatic image annotation using salient object detection and improved k-Nearest Neigbor classifier named ASALTAG. ASALTAG is consist of three major part, the segmentation using Minimum Barirer Salienct Region Segmentation, feature extraction using Block Truncation Algorithm, Gray Level Co-occurrence Matrix and Hu’ Moments, the last part is classification using improved k-Nearest Neigbor. As the result we get maximum accuracy of 79.56% with k=5, better than earlier research. It is because the saliency object detection we do before the feature extraction proccess give us more focused object in image to annotate. Normalization of the feature vector and the distance measure that we use in ASALTAG also improve the kNN classifier accuracy for labeling image.