Analysis of Facial Image by Gradient Boosting Locally Extracted Features
A. Vinay, Vinayaka R. Kamath, Yashvanth Kumar Guntupalli, K. N. Balasubramanya Murthy, Senthil Kumaran Vijayalakshmi Natarajan · 2019
Face recognition systems running on commodity hardware arguably play the most important role as reinforcement for human computer interfacing. This paper intends to tackle the specific hurdle of labeling the person present in the given facial image, pertaining to different situations such as varying illumination, distinctive expressions, and noisy background. A novel pipeline that contemplates to solve the given complication is presented using a comparative study. Locally extracted descriptors are subjected to dimensionality reduction to structure the feature set. Compaction of features is achieved through aggregators such as VLAD, Fischer vectors and Bag of LARK features. Aggregated vectors are classified using gradient boosting techniques. The stack is tested using multiple benchmark datasets with appropriate splits to cross validate the results. It was observed that the model performed at its best when there were changes in the facial expressions of the subjects. Comparative paradigm helped to accurately examine the several approaches proposed by the paper.