Cross-modal face recognition with illumination-invariant local discrete cosine transform binary pattern (LDCTBP)

With the ever-increasing security threats in recent years, biometric authentication has become omnipresent. Among all biometric characteristics, face recognition research has gained traction lately. This paper proposes a new face image descriptor named Local Discrete Cosine Transform Binary Pattern (LDCTBP) for illumination- and modality-invariant face recognition. Utilizing the frequency segregation behavior of Discrete Cosine Transform (DCT), an effective cross-modal illumination-agnostic local feature descriptor has been formulated. Eventually, by encoding the illumination-normalized DCT coefficients into a binary pattern, Local Discrete Cosine Transform Binary Pattern has been generated. Qualitative and quantitative analysis performed on the Extended Yale-B, CUFSF, and TUFTS dataset depict the supremacy of the proposed framework over other state-of-the-arts. Moreover, the proposed LDCTBP has been integrated with a light-weight Convolutional Neural Network (CNN) to prove the importance of handcrafted features in CNN training. Graphical abstract