A comparative analysis on the performance of different handcrafted descriptors over thermal and low resolution visible image dataset

In the past visible light face recognition has been considered as an authentic biometric trait and a lot of research has been done for performance enhancement of the same. But the impact of illumination variation over the recognition performance has been a challenging task to overcome when we take outdoor conditions into consideration whereas thermal infrared features are robust to the illumination variations as it focus on changes of temperature distribution on facial muscles and blood vessels. These temperature variations in itself creates a texture. The other major reason for the need of thermal trait is the night time surveillance when there is low or minimal amount of light thus is unable to illuminate the face. Thermal facial images is majorly used in several military operations as covert data acquisition carry challenges in its own ways and the visible light images sometimes fail miserably to fulfil the desired outcome due to its dependence on lighting and illumination variations. In this work we have done a comparative study on the performance of different handcrafted descriptors over thermal and visible images using different distance measures. The paper divides the proposed descriptors in 2 different sets i.e. lower order derivative based and higher order derivative based descriptors. We have compared the performance of different lower order derivative based descriptors over the higher order derivative based on different distance measures and tried to analyze the limitations and delimitations of handcrafted descriptors over multispectral range of images.