A comparative study on the effectiveness of using different similarity and dissimilarity measures for head pose estimation
Shefna Shareef · 2017
This paper presents a comparative study on the effectiveness of using some similarity and dissimilarity measures for head pose estimation. The aim is to find the most efficient measure for head pose estimation from a set of measures. The efficiency is determined based on the error each measure produces when a test image is compared against a set of trained images and the execution time taken by each measure. To ensure best results, some image enhancement techniques are used on the dataset and the appropriate enhancement technique is chosen by an image quality assessment method based on the gradient and luminance similarity. The measures are applied on images under different scenarios such as darklight condition, normal light condition and light condition. Using Wiener filter followed by histogram equalization as an enhancement technique gives best result for this pose estimation method. The experimental result showed that Pearson correlation coefficient and Spearman's Rho among the similarity measures and Li norm, MAD, Square L2norm, MSD among the dissimilarity measures can be effectively used for head pose estimation under any light condition.