Zero-crossing based image projections encoding for eye localization

This paper introduces a feature extraction and classification framework that is capable of fast and accurate description of the eye area. The extracted features are based on the combination of integral image projections with TESPAR signal encoding. While image projections were widely used in the field of image analysis, TESPAR was developed for speech recognition and it was only very recently used in conjunction with image signals. Using the unique combinations of the two techniques, we construct features that are easy to compute and provide independency with scale and illumination variation. The computed features are used as inputs into a Multi Layer Perceptron and the capabilities of the resulting framework are proved in an application of eye localization.