To Construct Block-based Bag of Visual Words for Image Retrieving and Classification

• Main Authors: Wan-Hsin Li, 李宛馨
• Other Authors: Chung-Ming Kuo
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/63424594694834691255

碩士 === 義守大學 === 資訊工程學系 === 101 === Nowadays, multimedia applications have become very popular in our life. One important issue is to develop an effectively image retrieval system to manage enormous multimedia database. Content-based image retrieval can extract features accurately, and therefore has been widely used in image retrieval.In this thesis, we first select non-overlapped blocks as our training set from image database randomly. Each 4×4 block is categorized as macro block or micro block according to block information. Then, the blocks are trained and merged by comparing their similarity to form the representative blocks. The trained micro and macro blocks are collected and encoded into micro and macro codebook, respectively. The encoded histograms are used to measure their similarity between images. Experimental results indicate that the proposed method achieves high retrieval performance.