Generating images of partial face using landmark based k-nearest neighbor

Abstract

One of the most common approaches to address the partial face recognition challenge is to crop the full face image into segments. The problem is how the full face image must be cropped in a uniform way to generate informative segments. The un-blindly strategy was applied in this paper to generate informative segments, it depends on localizing the facial landmarks and selecting the more informative facial points as a key points, as more as the k-nearest neighbor concept was explored to select the k nearest landmark points to the key points. Two landmark localization techniques were experimented, the suitable technique resulted in segments which are overlapped due to the supervised clustering technique that explored in this paper to cover important biometric face regions, not repeated and covered most probabilities in which it is possible to distinguish the query face from the available part of it.