Current Biases and Future Research Avenues for Archive Image Colourisation

Recent advances in deep learning (DL) have resulted in a proliferation of techniques able to colourise achromatic or monotone images. This was a previously dormant area of image processing due to the under-constrained nature of the problem. Many of these novel methods have been documented and reviewed. Despite this, many colourisation problems persist that include chromatic aberrations, lack of standard datasets and the absence of a quality no-reference assessment measure. This paper presents a brief review of the research field and highlights these salient problems re-framing them as possible avenues for further research; particularly in the archive image domain.