VISUAL SEARCH(*) OPTIMIZATION FOR NEAR DUPLICATE IMAGES

• Main Author: Sukma Limanus, Steven
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51307
• Institution: Institut Teknologi Bandung

This theseis talks about visual search(*) optimization for near duplicate images. Visual search(*) mentioned is visual search implementation at eBay as explained by Yan, et al with limitation on the use of binary hashing to do visual search. This thesis explore optimization with that limitation. This thesis tries to calculate near duplicate based on hamming distance as implemented by Lin, et al based on the pre calculated binary hash implemented by Yan, et al. Furthermore this thesis also try to calculate near duplicate by modifying Hu, et al implementation by removing the convolutional layer and use binary hash as the input instead of image. The result of this thesis shows that the use of both approach to detect near duplicate is able to give the expected result. Furtheermore, the hamming distance approach needs shorter time for indexing