Fast full-search schemes for pattern matching using integral images

Abstract

Pattern matching plays an important role in various fields such as image registration, stereo matching, image coding, texture synthesis, image retrieval, object tracking, etc. Since the pattern matching process have been a time-consuming work, many fast pattern matching algorithms have been proposed. They can be categorized into several groups: 1) candidate sampling; 2) simplification of matching criterion; 3) bitwidth reduction; 4) hierarchical search; and 5) fast full-search. By adopting mathematical inequality, fast full-search can be boosted while preserving the same result of full search algorithm. In this dissertation, fast full-search pattern matching schemes based on four different criteria (sum of absolute difference, sum of squared error, normalized cross correlation, and pixel difference classification) are further investigated and compared to several conventional schemes. Key tools of proposed schemes are the Minkowski inequality, the Cauchy-Schwarz inequality, and the concept of integral image. Simulation results show that proposed schemes outperform conventional schemes.