Matching Edges in Images ; Application to Face Recognition

TL;DR

This paper introduces a method for image comparison based on edge representation and geometric basis search, enabling face recognition by matching edge patterns despite transformations and detection errors.

Contribution

It presents a novel edge-based image similarity measure using geometric basis search, applicable to face recognition and pattern matching.

Findings

Effective in face recognition despite transformations

Requires only a few tens of trials for reliable matching

Applicable to various pattern matching problems

Abstract

This communication describes a representation of images as a set of edges characterized by their position and orientation. This representation allows the comparison of two images and the computation of their similarity. The first step in this computation of similarity is the seach of a geometrical basis of the two dimensional space where the two images are represented simultaneously after transformation of one of them. Presently, this simultaneous representation takes into account a shift and a scaling ; it may be extended to rotations or other global geometrical transformations. An elementary probabilistic computation shows that a sufficient but not excessive number of trials (a few tens) ensures that the exhibition of this common basis is guaranteed in spite of possible errors in the detection of edges. When this first step is performed, the search of similarity between the two images reduces to counting the coincidence of edges in the two images. The approach may be applied to many problems of pattern matching ; it was checked on face recognition.