Supplemental Material for User-guided White Balance for Mixed Lighting Conditions

1 Importance of the single-light white balance step The first step of our pipeline involves a standard single-light white balance iteration. This comes from our observations that after a single-light white balance correction we usually have a few natural looking regions. By fixing those regions, using the correct-color scribbles, users provide important boundary conditions for the later optimization. In this section we study how the final result of our system depends on the initial single-light white balance step. Note that we use the same set of scribbles (the one given in the main paper), what changes for each result is the input image on which the scribbles act. (a) Input (original) (b) Single-light wb (temperature = 3000) (c) Single-light wb (temperature = 2600) (d) Our result (no single-light wb) (e) Our result (warmer) (f) Our result (cooler) Figure 1: Placing the scribbles on the original image (a) affects the boundary conditions of our system. Although this leads to propagation of non-natural looking colors, our optimization is still able to remove unpleasent color casts along the way, image (d). However, the mixed lighting is observed through the unnatural looking colors (e.g. the bricks wall, the drawers). Next, we show comparison with two different single-light white balance settings. In the first one, image (b), we apply single-light white balance correction with higher temperature (3000K). This produces image (b), that has warmer colors. In the second one, we use white balance correction with lower temperature (2600K), which gives cooler result, image (c). Our optimization respects the initial data and correspondingly produces warmer result, image (e), and cooler result, image (f). In all cases, the results of our mixed-lighting white balance system looks preferable to the input image.