Near duplicate image identification

Abstract

Near duplicated images refer to a pair of images in which one is close to the exact

duplicate of the other, but different in the capturing conditions, times, rendering

conditions, and/or editing operations.

Near Duplicate Image Identification (NDII) is a computer vision problem. It is more

challenging to identifying near duplicate images with difference in variation and

scale. Spatially Aligned Pyramid Matching (SAPM), a multi-level matching method

which is proposed by D. Xu el at robustly handles spatial shifts as well as scale

changes.

SAPM was implemented and experiment was done to test on effectiveness of SAPM

on NDII. The result clearly shows that SAPM is capable in identifying duplicate image.

A prototype was created to illustrate the use of SAPM in NDII.

To Improve SAPM efficiency and effectiveness, different features detector and other

clustering method can be tested or analysed. Near Duplicate Video Identification

(NDVI) can be implemented with SAPM and recently proposed video matching

framework Spatio-Temporal Pyramid Matching (STPM) could also be researched on

and implemented for NDVI.