Correction to: Illumination-aware group portrait compositor (The Visual Computer, (2022), 38, 12, (4009-4018), 10.1007/s00371-022-02508-z)

The publication of this article unfortunately contained mistakes. Figure 1 and the legend of Fig. 3 were not correct. The corrected Fig. 1 and legend of Fig. 3 are given below. (Figure presented.) Composite group portraits. The group portraits were produced from five individual portraits shown in the column Sources, each of which was captured in a different environment. Standard composite shows the composite results without considering visual coherence. Proposed shows the composite results using the proposed framework. Reference shows the rendered images of the 3D reference models. RMSE visualizes the root mean square error for each pixel between Proposed and Reference with a Turbo colormap [2], which considers visual continuity. The composite images without considering visual coherence have neither a uniform color tone among the elements nor cast shadows, while the composite images produced by the proposed framework compare favorably with the rendered one. Further, the upper portrait is configured with a simple background scene and the visual effects between the people and the scene can be confirmed clearly, while the lower one is configured with a photorealistic background scene and it evokes actual usage scenarios Fig. 3 Low-quality photometric information leads to poor composite results. The composite result that uses the estimated photometric information is compared with the other composite result that uses the ground truth photometric information. This preliminary experiment shows that even if the highest-level methods currently available are applied, the photometric information for obtaining a satisfied composite result cannot be estimated. U²-net [25], IIW [4], and pix2pixHD [36] The original article has been corrected.