Perceptual Evaluation of Demosaicing Artefacts

Abstract

Most of the digital camera sensors are equipped with the Colour Filter Arrays (CFAs) that split the light into the red, green, and blue colour components. Every photodiode in the sensor is capable to register only one of these components. The demosaicing techniques were developed to fill the missing values, however, they distort a scene data and introduce artefacts in images. In this work we propose a novel evaluation technique which judge a perceptual visibility of the demosaicing artefacts rather than compares images based on typical mathematically-based metrics, like MSE or PSNR. We conduct subjective experiments in which people manually mark the visible local artefacts. Then, the detection map averaged over a number of observers and scenes is compared with results generated by the objective image quality metrics. This procedure judges the efficiency of these automatic metrics and reveals that the HDR-VDP-2 metric outperforms SSIM, S-CIELAB, and also MSE in evaluation of the demosaicing artefacts.