Implementation of Robust SIFT-C Technique for Image Classification

Abstract

This paper describes the development of a robust technique for image classification using scale invariant feature transform (SIFT), abbreviated as SIFT-C. The proposed SIFT-C technique was developed to cater for varying conditions such as lightings, resolution and target range which are known to affect classification accuracies. In this study, the SIFT algorithm is used to extract a set of feature vectors to represent the image and the extracted feature sets are then used for classification of two classes of weed. The weeds are classified as either broad or narrow weed type and the decision will be used in the control strategy of weed infestation in palm oil plantations. The effectiveness of the robust SIFT-C technique was put to test using offline weed images that were captured under various conditions which truly reflect the actual field conditions. A classification accuracy of 95.7% was recorded which implies the effectiveness of the SIFT-C.