Estimation of Optical See-Through Displays

Optical See-Through displays (OST Displays) require the appearance of the virtual content to closely resemble that of the real world. However, current OST displays typically show color distortion due to hardware limitations. To overcome this problem, we propose a chromaticity based local linear regression model for color distortion estimation of OST displays. From our initial experiments, it was observed that local samples with similar chromaticity can be modeled more linearly rather than simple nearest neighbors in a color space. We argue that the accuracy of local linear regression can be further improved by placing emphasis on similarity of chromaticity. Experimental results showed that the proposed chromaticity based color correction achieves higher accuracy compared to the nearest neighbor method, for any training set size or neighborhood size as well as global regression. It confirms that chromaticity based nearest neighbors can be modeled more linearly together than simple nearest neighbors in a color space.