Image colourisation as a compression technique

Abstract

Digital image colourisation is a topic within the domain of image processing which deals with automatic processes that apply colour to grayscale image. One appreciates that this is a non-trivial process as there are often no clues as to the colours used in the original image. In fact, many image colourisation algorithms require some form of user assistance, usually an artist or designer, who supplies additional colour information as hints to the colourisation algorithm. Image compression is another ongoing active field of research. JPEG has been a canonical compression technique for over a decade now and there are many attempts to design a better compression technique than JPEG. Colourisation encoding is a combination of these two topics such that colourisation techniques are used as a means for image compression. This line of research is called colourisation coding and is the primary topic for this project. In this report, we look at various recent attempts at image colourisation, image compression and colourisation coding followed by our proposal of an algorithm as an alternative to existing colourisation coding processes. Our algorithm is original in that it has a unique prefiltering step which allows it to make minor modifications to the underlying image so that the colourisation techniques we employ are more accurate. Briefly, our algorithm takes a colour image and converts it to its grayscale equivalent. The grayscale image is derived using the original image's luminance channel and then prefiltered as necessary. The resultant grayscale image is used to segment the image into a number of regions such that each region is assigned a colour corresponding to whatever colour was in the original image. Since the segmentation process is deterministic, the encoding only stores a grayscale image and a list of colour chrominance values. A decoder is then able to undergo the same segmentation using the supplied grayscale image and restore the chrominances to each segment according to the chrominance list given in the encoding. In this report, we explain in detail the design and implementation of this algorithm. We thoroughly evaluate and discuss the performance of the algorithm using a set of criteria and the evaluation is then followed by a number of suggested improvements to the algorithm. The results show that there is indeed potential in our algorithm. Hence, we also propose a number of topics for further research in the future based on the work we present for this project.