Re-ranking for web image search

• Main Author: Cheng, You Jun
• Other Authors: Xu Dong
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: http://hdl.handle.net/10356/59131
• Institution: Nanyang Technological University

With the rise of smartphone enabled camera, people are taking more photos because of the ease of convenience of pointing and shooting. With more photos being taken, there would be an increase in image uploads on image sharing website which causes searching for relevant images in this large database of images more difficult. In this project, the aim is to create a new image re-ranking framework based on Text-based information retrieval (TBIR). The proposed framework works in 3 stages, the initial ranking stage, Weak Bag Annotation Stage and the SVM training and prediction stage. In the initial ranking stage, images are given a ranking score based on their surrounding tags and the position of the tag that the search query matches. In the weak bag annotation stage, images are clustered together based on their textual and visual feature. The clusters are treated as bags where only the bags are labelled while the images within the bags are unknown hence this is a weak bag annotation. Using this bags, the SVM model is trained and using the model, a label is given to each image of the queried images based on the prediction of the model. The images are then finally re-ranked based on their ranking score. The experiments will be conducted on the real world NUS-WIDE dataset to display the effectiveness of the proposed framework.