Shape-based Image Retrieval

Shape-based Image Retrieval By Aditya Vailaya Retrieval e ciency and accuracy are two important issues in designing a contentbased database retrieval system. We propose a new image database retrieval method based on shape information. This system achieves both the desired e ciency and accuracy using a two-stage hierarchy: in the rst stage, simple and easily computable statistical shape features are used to quickly browse through the database to generate a moderate number of plausible retrievals; in the second stage, the outputs from the rst stage are screened using a deformable template matching process to discard spurious matches. We have tested the algorithm using hand drawn queries on a trademark database containing 1; 100 images. Each retrieval takes a reasonable amount of computation time ( 45 seconds). The top most retrieved image by the system agrees with that obtained by human subjects, but there are signi cant differences between the top 10 retrieved images by our system and that provided by human subjects. This demonstrates the need for developing shape features that are better able to capture human perceptual similarity of shapes. An improved heuristic has been suggested for more accurate retrievals. The system matches lled-in query images against lled-in images from the database thus using only the gross details in the image. Experiments have shown that matching on the lled-in database extracts more images that have similar content. The proposed scheme has also been tested on a database of master-marks of Norwegian silversmiths containing 171 images. The experiments show the e ectiveness of the integration of shape features for matching faded, oriented, and scaled master-marks extracted from the bottom of the silver tankards. We believe that it is a challenge to develop an automatic retrieval algorithm which models the human perception well. However, considering the time and e ort for humans to perform such a retrieval task, an automatic retrieval algorithm can provide a powerful tool to facilitate the retrieval task for human beings. Acknowledgments I would like to acknowledge all the people who have assisted me throughout my studies at Michigan State University. I am extremely grateful to my advisor, Professor Anil Jain, for his continuous guidance and encouragement, and the valuable time that he spent on this project with me. I owe my great interest in research to him. I am extremely grateful to Dr. Ming-Yee Chiu and the technical sta at Siemens Corporate Research, Princeton for their guidance and encouragement during my stay there in 1995 summer and the use of their database for this research project. I would also like to acknowledge Dr. Britt Kroepelien for providing a strong basis for the project on authentication of Norwegian silver tankards. It was with her patience and determination that we were able to start this project in such a short time. I would like to specially acknowledge Chitra Dorai for being my mentor in the initial period of my stay here and for patiently assisting me to adjust to the new environments. I would like to dedicate this thesis to my family who have constantly encouraged me to achieve new heights. Without their love, patience, understanding, and support, I would not have nished this thesis. iv I would like to thank Yu Zhong for her assistance in using deformable template matching. I would also like to acknowledge the help and support I have received from members of the PRIP lab and from my friends at Michigan State University.