#TITLE_ALTERNATIVE#

• Main Author: Rahman (NIM 23505014), Arif
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/7292
• Institution: Institut Teknologi Bandung

Abstract: <br /> <br /> <br /> <br /> <br /> Content-Based Image Retrieval (CBIR) system finds images based on similarity of their visual content, i.e color, texture or shape. CBIR involve collection of documents with huge number, that is for retrieval efficiency, index of visual content features should be built. A common indexing method in CBIR is multidimensional -index using tree structure. <br /> <br /> <br /> <br /> <br /> A CBIR system using fractal codes of image document, called SPECKTRAL is proposed in this research. SPECKTRAL is analyzed and designed using Unified Process method and implemented in web-based environment using Java Servlet. Fractal code is code from image encoding process in FIC (Fractal Image Compression). Main characteristic of FIC is self-similarity searching between image parts based on value of pixels intensity. From that characteristic, can be assumed that similarity of fractal code represents similarity of pixels intensity variance pattern which visually represents the image texture. Similarity of fractal code measured with average of minimum Euclidean distance difference between contrast and brightness coefficients in fractal code. Index structure that used in this system is Inverted File which consists of fractal codes and pointers to document which that codes appear. An Inverted File is used because it has lower computation complexity than multidimensional index. <br /> <br /> <br /> <br /> <br /> The experiment result shows that similarity of fractal code between two images represents similarity of their textures. These could see from the query testing result toward Brodatz and Vistex image collection which is in average, 64,61percent from all the image with similar texture rank on the top.