Sparsity based facial region detection from minimal training data

In recent times sparse framework based signal modelling has been extensively used in various signal processing tasks. State-of-the-art results have been obtained using this approach in various image processing applications. In this paper, we have adopted the sparse framework for the task of detection of various facial regions such as eyes, lips, nose etc. We propose a scheme for modelling these regions by signatures using dictionary learning in the sparse framework. The proposed algorithm has been tested on FEI face database. Experimental results show that the proposed scheme is robust and detects different regions with very high accuracy. Although, there exists a number of machine learning algorithms for this task, proposed algorithm is novel in the manner that it uses very few training images and is suitable for applications where inadequate sample data is available for training.