MIFT: A framework for feature descriptors to be mirror reflection invariant

摘要

Visual matching is one of the most fundamental and important tasks in computer vision and pattern recognition. The images often appear to be scaled, rotated, view point changed and flipped (mirror reflected). The popular way to match such images employs local image features, such as SIFT and its variations (e.g. OpponentSIFT and HSVSIFT). Although the common local image features are able to deal with the transformations including scale, rotation and view point change, they fail to handle the mirror reflection. This paper presents a framework, mirror-reflection invariant feature transform (MIFT), for improving their degenerated performance due to mirror reflections. The experiments demonstrate the robust performance of MIFT. An application to symmetric axis detection is also shown.