A Robust Illumination-Invariant Face Recognition Based on Fusion of Thermal IR, Maximum Filter and Visible Image

Abstract

Face recognition has many challenges especially in real life detection, whereby to maintain consistency in getting an accurate recognition is almost impossible. Even for well-established state-of-the-art algorithms or methods will produce low accuracy in recognition if it was conducted under poor or bad lighting. To create a more robust face recognition with illumination invariant, this paper proposed an algorithm using a triple fusion approach. We are also implementing a hybrid method that combines the active approach by implementing thermal infrared imaging and also the passive approach of Maximum Filter and visual image. These approaches allow us to improve the image pre-processing as well as feature extraction and face detection, even if we capture a person’s face image in total darkness. In our experiment, Extended Yale B database are tested with Maximum Filter and compared against other state-of-the-filters. We have conduct-ed several experiments on mid-wave and long-wave thermal Infrared performance during pre-processing and saw that it is capable to improve recognition beyond what meets the eye. In our experiment, we found out that PCA eigenface cannot be produced in a poor or bad illumination. Mid-wave thermal creates the heat signature in the body and the Maximum Filter maintains the fine edges that are easily used by any classifiers such as SVM, OpenCV or even kNN together with Euclidian distance to perform face recognition. These configurations have been assembled for a face recognition portable robust system and the result showed that creating fusion between these processed image illumination invariants during preprocessing show far better results than just using visible image, thermal image or maximum filtered image separately.