Abstract
The aim of this study is to provide an automatic method for the interpretation of images of objects that are coated with thermal paint. Thermal paint changes colour permanently according to the temperature to which it is heated and can be employed as a temperature gauge where more cumbersome measurement apparatus may not be suitable. Such a gauge requires a means to convert the manifestation of the measurement to the corresponding numerical values. In our case this involves the grouping of ranges of colour together into temperature bands and the extraction of the temperature contours between these bands, a task currently performed by a human operator. This study will demonstrate some success in the automatic interpretation of thermal paints through computer vision approaches. In summary the main contributions of this work are: The demonstration that edge detection is not a useful step. Human operators tend to interpret thermochromic paint not simply by colour matching, but by locating prominent colour change points. We demonstrate why in our opinion this in not necessarily the best step through an exploration of colour edge detection. The development of a feature space model of the paint colour formation based on B-splines and the employment of this within a maximum likelihood estimation scheme [GKWG96],[CSGW97] The development of a paint interpretation method based on a Markov Random Field and Simulated Annealing [GSW+98] Our methods axe applicable to cases of ideal data. We highlight some troublesome paint artefacts that occur in real cases and that hinder interpretation. We discuss possible solutions. Finally we draw conclusions and point to directions for possible future work. Key words: thermochromic paint, maximum likelihood estimate, simulated annealing.