Fuzzy Rule-Based Sistem Temukembali Citra Bunga

ADI SUCIPTO AJI. Fuzzy Rule-Based Image Retrieval Systems. Under the direction of MARIMIN and YENI HERDIYENI. The common problems on image retrieval systems are vagueness and ambiguity of human perception of image similarity and rigidness on weighted features. This research develop a new method for measuring image similarity base on color and shape features by embedding the fuzzy logic called fuzzy rule-based method. Fuzzy rule-based as human thinking representation tends to capture systems user subjectivity. Two major processes in image retrieval systems are indexing and retrieval. Indexing process are image features segmentation process by color using histogram and shape using invariant moment. Retrieval process are fuzzy logic implementation with following steps : image features fuzzyfication, inferences among fuzzy rule-based, defuzzyfication and visualisation of relevant image nomination. Fuzzy rule-based for measuring image similarity comprise three linguistic variables sama, mirip and beda on each image features. The experiment result shows that implication method influences the systems performance. Highest precision values on Mamdani, Algebra and Einstein methods respectively are 86.44 %, 87.89 % and 87.56 %, which is by perception criteria Mamdani’s method producing better image than the others. Highest performance of fuzzy rule-based on 3 implication method are : if color is sama or beda and shape is sama or mirip or beda then image is sama or beda depend on color classification ; if color is mirip and shape is sama or mirip then image is mirip ; if color is mirip and shape is beda then image is beda.