Preferred white balance for applications using virtual backgrounds

Abstract

AbstractUnder the current COVID‐19 global pandemic, most of the world is operating online, which has increased the importance of better understanding the perceived color quality of video conference calls. We performed two experiments to evaluate the white balance appearance preference for images simulating a scene from a video conference call where a person is using a virtual background. Due to the dissimilarities of the light sources used for the subject and the background scene, the overall picture may look aesthetically unappealing. The first experiment was designed to assess the preference of white balance for images containing a foreground subject, with three different skin tones: light, medium, and dark, and a background scene, with five different color temperature appearances, cool to warm. The background scenes include famous attractions in the United States, water bodies, foliage, and a few less common scenes like a sculpture at the Rochester Institute of Technology (RIT) and a ColorChecker. Observers were presented with a pair of images of the same subject with same background scene, but with different white balance appearance. These comparisons were performed for each foreground subject with all background scenes. Both experiments were performed by naïve observers, who were from around the globe with no knowledge of color science and observers from the Munsell Color Science Lab (MCSL) at the RIT. The results show that observers' preference increases as we go from cooler to warmer appearance for the Canyon, RIT, Snow, Flower, and Autumn backgrounds, and vice versa for the Lake background. The Golden Gate background results are the most scattered among all scenes with very small differences in their scale values. MCSL observers show a strong agreement in preference to warmer appearance for light and dark skin tones for the RIT scene, and neutral and warmer appearance for medium skin tone for the RIT and Golden Gate scenes, respectively. To understand the relation between the background scene and foreground subject's white balance appearance, a second experiment was performed. The subjects had three different appearances: cool, neutral, and warm. Based on the results of the first experiment, five scenes with two white balance appearances, the most preferred rendition and the neutral rendition (original appearance at the time of the scene was captured), were used for the second paired comparison study. The results indicated that preference varied based on the foreground subject skin tone and background scene. However, the background preferences follow similar trends to the first experiment. There were variations in the degree of agreement, some showed very strong agreement, like dark skin tone with the RIT background, whereas for others, like autumn background with light skin tone, preferences were more scattered. Overall, the differences in the scale values were smaller as compared to the first experiment, which indicates that, as we present the observers with more options, the decision became harder.