ABSTRAKSI: Riset yang berhubungan tentang Image Annotation and Retrieval telah sangat berkembang saat ini. Dimulai dengan mimpi tentang bagaimana cara untuk mengorganisasikan sekumpulan citra skala besar tanpa melihat dulu isi citra tersebut, dan pada tahun 90 an muncul ide untuk mengorganisasikan citra dengan melihat isi dari citra tersebut atau lebih sering disebut dengan Content-Based Image Retrieval (CBIR). Salah satu metode baru yang dapat digunakan untuk meretrieve citra adalah Supervised Learning of Semantic Classes. Dalam membentuk model matematis, Supervised Learning standar menggunakan Gaussian Mixture Model dan Expectation Maximaliztion untuk Maximum Likelihood Estimationnya.
Dalam tugas akhir ini, penulis berusaha mengganti model matematis pada Supervised Learning tersebut menggunakan Generalized Gaussian Mixture Model untuk mixture model-nya dan Split Merge Expectation Maximaliztion untuk Maximum Likelihood Estimation-nya.
Berdasarkan hasil uji, secara umum metode Supervised Learning dengan GGMM-SMEM menghasilkan citra retrieve yang lebih akurat dibanding dengan menggunakan GMM-EMKata Kunci : content based image retrieval, image annotation, image retrieval, supervised learning of semantic classes, gaussian mixture model, generalized gaussian mixture model, expectation maximalization, split merge expectation maximalization, maximum likelihood eABSTRACT: Research about Image Annotation and Retrieval now is more developed. Starting from dream about how to organize a group of large scale images without knowing it's content and at early 90's there is an idea to organize images with knowing it's content, it's call Content-Based Image Retrieval (CBIR). One of new method that can use to retrieve image is Supervised Learning of Sematic Classes. In making mathematic models, Supervised Learning standard using Gaussian Mixture Model for mixture model and Expectation Maximaliztion for Maximum Likelihood Estimation.
In this paper, author want to change how to make the mathematic models in Supervised Learning with Generalized Gaussian Mixture Model and Split Merge Expectation Maximaliztion for Maximum Likelihood Estimation.
Based on result, generaly SML method with GGMM-SMEM retrieve images more accurate than SML method with GMM-EMKeyword: content based image retrieval, image annotation, image retrieval, supervised learning of semantic classes, gaussian mixture model, generalized gaussian mixture model, expectation maximalization, split merge expectation maximalization, maximum likelihood e