An improved local pattern descriptor for biometrics face encoding: a LC-LBP approach toward face identification

摘要

Local binary pattern (LBP) has proved to be an efficient local image descriptor used in computer vision. LBP operator has successfully been applied in many biometrics applications. Face recognition is one such application where performance degrades due to varying illumination, facial expressions and head pose changes. To address these issues effectively, LBP operator can be used to represent face images during feature extraction. However, due to the limited number of discriminative characteristics, the LBP operator cannot address these issues properly. It seeks more distinctive and well-structured local feature representation. This paper reports a face identification system which makes use of a novel LBP variant called Logically Concatenated LBP in which the basic notion of Modified LBP is exploited and combined with three logical operations, viz. AND, OR and XOR. These logical operations are used to produce three different labeled face patterns from a single face. Further, these patterns are used for face matching and identification. The proposed face identification is tested on FEI, UMIST and Extended Yale Face B databases and identification accuracies have been determined. Experimental results are found to be encouraging and convincing. The experimental results are also compared with some existing local descriptors.