Optimal Colourization For Greyscale Images

摘要

This paper proposes a user friendly interactive colourization system that utilizes iterative region expansion and linear programming to carry out colourization. The region corresponding to each scribbled colour is iteratively expanded using a sequence of increasing threshold values. Usually a few pixels remain unassigned to any colour region after this stage, each of which is then simply assigned to the nearest region based on an intrinsic distance measure. The optimal colours for the brightest pixel and the darkest pixel in each region are determined by employing linear programming, and then the whole region is colourized using linear interpolation. The proposed system provides several colourization modes users can select from. The user simply needs to scribble on the given greyscale image and select the colouring mode they want. The experimental results show that the proposed system provides natural and satisfactory colouring results.