We compute the difference d(T,C) between canvas image C and target image T and select the regions with sufficiently high image differences for the given color i. These regions, we call them di(T,C), are our candidate regions for adding new brush strokes.