Simulation of greyscale image colouring using blob detection

Abstract

Automatic colouring of greyscale images using computer is one of the important fields in digital image processing. It helps to produce more appealing visuals to human eye when one have to deal with medical images, night vision cameras or scientific illustrations. However, to produce images that are at par with the ability of human eyes, computerised colouring process takes a lot of time and ample calculation. Recent years, blob detection has shown a good development for finding features in an image. This method not only can run on low memory devices but also provides users with faster calculation. Encouraged by these advantages – work on low memory devices and enable faster calculation, two models of untrained colouring of greyscale images are proposed in this study. The maximum number of blob features is examined using Centre Surround Extremas (CenSurE) and Binary Robust Independent Elementary Features (BRIEF). The result of this study proves that the images coloured by these models look better with increment features of the key point if the minimum matching distance is as low as possible. In addition, when comparing feature descriptors using Fast Retina Keypoint (FREAK) solely and FREAK together with Speeded-Up Robust Features (SURF), it is concluded that the result is getting better with the decrement of minimum Hessian in the image. This experiment leads to the discovery that the selection of feature descriptors will influence the result of colouring.