Most machine vision research is performed using grey-scale images. There are three main reasons for this. The first is practical: cheap colour cameras work by using a grid of red, green and blue filters arranged over the detector. Therefore, the RGB components of any given pixel were generated from slightly different spatial positions: this spatial error complicates any subsequent analysis. Cameras exist that contain optics to split incoming light into three separate paths, passing through red, green and blue filters and falling on separate detectors: professional television cameras are one example. However, these cameras tend to be much more expensive (a factor of 400 is typical) and therefore fall outside the budgets of most researchers. The second reason is theoretical: whilst it is simple to define an algebra for grey-scale images (as represented by the Imcalc tool in Tina), it is much more difficult to define a satisfactory equivalent for colour images. The third reason is biological: humans can operate perfectly well in a grey-scale environment, for instance watching a black-and-white movie, implying that colour perception is not essential in order to perform the majority of basic visual tasks. For these reasons, colour-based algorithmic functionality in Tina is limited.
There is one exception to the above: the three colour channels of an image can be treated as independent images, and data fusion performed for processes such as segmentation. Therefore, whilst the infrastructure of Tina currently provides little support for colour images (for example there is no colour Tv tool, or colour image structure) the Tina Colour Image tool provides support for input of colour images as a separate RGB fields, conversions to a number of other colour space, and colour segmentation by non-parametric density estimation in colour space, as described in Tina Memo no. 2001-015.