Images Classification Using SIFT And RANSAC Method

Abstract

Taking pictures has become common thing to do for capturing certain places. The process of taking and storing pictures that quite easy affecting the number of pictures taken and not well arranged. This results in difficulty finding desired image with ease. Several applications have been created to facilitate the process of organizing images. Generally, the applications use metadata and geotagging of the images. However, metadata and geotagging cannot be seen by naked eye in the picture.

So in this paper the application that classifies images based on object which appear in the images will be made. To make this application, SIFT feature from image will be used. Additionally RANSAC method will be used to improve the accuracy of matching SIFT features. A database is also set up to avoid ambiguity in matching SIFT features and speed up the process of grouping.

Based on the conditions at the time of testing the average value of success rate of classification with SIFT and RANSAC method can reach 72.5%. In addition it was found that SIFT features are vulnerable to the view point, tilt, non-linear illumination and lighting effects due to texture objects. Thus providing patterns that can cover the changes on SIFT features is necessary. In addition, the image preprocessing is needed to eliminate the possibility of ambiguity pattern matching on SIFT features due to background, texture, and patterns which contained in images.