The idea of the innovation project RE:Color addresses the large amount of black and white (b/w) films stored in the archives of this world. Today, these historical cultural assets are preserved from decay by digitization and (sometimes) improved by digital restoration. However, the marketing of such b/w films is difficult due to limited public acceptance.
Although the first efforts in coloring b/w films date back to the 1970s and 1980s, there was little progress on the topic until the very last years. Main reason was the tremendous effort and related costs in manual or semi-automated colorization techniques. The rise of most recent artificial intelligence and deep learning technologies lead to a high number of scientific publications. However, the practical usability of such publications in real industrial environment has not been proved yet. So far there is no single commercial off-the-shelf solution available.
The proposed project RE:Color uses a multilateral approach and tries to merge brand-new scientific approaches into one singe environment. The result is an integrated software application that combines interactive and automated colorization techniques with deep learning technologies in order to reach a predominantly automatic but still full user-controlled colorization process. This central requirement can only be achieved, if the neuronal self-learning networks are well trained and can be dynamically influenced by user interaction.
The main innovation of the planned project lies in the adaptation of different academic developments and their re-use and bundling into a single workflow, allowing efficient and fully user controlled colorization processes, based on historic proven colors and models.
The success of such a solution on the market is highly probable due to the increase in efficiency.