PENGUKURAN KEMIRIPAN-SEMANTIK CITRA DENGAN ~-' ~-'-"- . -- 'BAG OF WORDS'LOKAL FRAGMEN"--SEMANTIK CITRA ~ -'- ", -- ----- ---- ~,,,,-,~,·--·.'._w_. __ ,._.... ~ . .,,,/ IMAGE SEMANTIC SIMILARITY MEASUREMENT USING 'BAG OF WORDS' LOCAL IMAGE FRAGMENT

Abstract

This research proposes a new approach for measuring semantic image similarities between visual image using 'bagof- words' semantic local image fragments. In this approach, image representation is represented using an ordered list of local image fragment. Local image fragment is a part from images which gives a significant information and appears consistently within images. Local image fragment was extracted using online learning. Image semantic similarity is measured by J) solving a correspondence problem between two ordered sets offeature; and 2) Calculating similarities between matchedfeatures and dissimilarities between unmatchedfeatures. The methodology has already been tested on several benchmark image databases containing 2,268 images. The experimental results show that the proposed method can increase precision by as much as 74% in average. The proposed method ispromising that could be used to measure semantic similarity adequately to human visual similar perception.