Automatic color realism enhancement for computer generated images

Photorealism has been one of the essential elements in producing computer generated imagery. The state-of-the-art techniques employ various rendering algorithms to simulate physically accurate light transport for generating a photorealistic appearance of scene. However, they require a labor-intensive tone mapping and color tunes by an experienced artist. In this paper, we propose an automatic photorealism enhancement algorithm by manipulating the color distribution of graphics so to match with that of real photographs. Our hypothesis is that photorealism is highly correlated with the frequency of color occurrence in real photographs; more often we observe more realistic we believe. Based on this hypothesis, we find principal color components by following two steps. First, we extract the most representative features from the color distribution of photographs. Then, we obtain the coefficients of the most distinguishable principal axis to separate the features of photographs and those of graphics. The distribution of these coefficients constructs the color distribution of graphics and real photographs, respectively. Then, we modify the statistical characteristics (orientation, variation and the mean of color distribution) of graphics according to that of photographs. Experiments and user study have confirmed the effectiveness of proposed method.

Figure optionsDownload high-quality image (315 K)Download as PowerPoint slideHighlights
► It is the first automatic photorealism enhancement technique using data-driven color priors.
► It can be applied onto the rendered frame for increasing the photorealism.
► It can be expanded into other classification methods, facial color or the lesions classification.