A Robust Multiple Adaptive Derivative Face Recognition System on Pose and Illumination

Computer system using CNN visions has proven to be flawed as it fails to capture all the different angles and positions of the image. The main problem with CNN is that it has is no proper system in localizing a person’s head without depending on the person image appearance. CNN based models can memorise images that are augmented and linked together after trained them with supervised positive and negative from the ground truth annotated images marked with landmarks but does not conduct face localization. Thus, the objective of this paper is to investigate, design a face recognition system that is able to recognise a person regardless of his or her head poses. This paper’s task is to investigate a proper system that will be able to be classify face localization. First, MAD Passive Image Processing (MADPIP) is used to perform restoring and enhancing all images. Second, MAD Face Detection (MADFD) is used to process face image with the optimization of Zhu Ramanan (OZR) algorithm to locate landmarks position accurately. Then, MAD Localization and Landmarking (MADLL) is used to build OZR model by performing landmarking classification. Next, the Optimized Zhu Ramanan (OZR) is designed to improve Zhu Ramanan (ZR) processing time by replacing HOG with OZR and finally all the extracted features are normalized. The results showed that the performance of the proposed Multiple adaptive derivative Face Recognition System (MADFRS)is acceptable in which the proposed MADFRS approach proposed a solution with accuracy percentage of 99.80% which is the third highest for LFW Dataset Benchmark.