Determination of the optimal colour space for distinguishing small retinal haemorrhages from dust artefacts

Abstract

AbstractPurpose To determine the optimal colour space from eight types of colour spaces for the purpose of distinguishing small retinal haemorrhages from dust artefacts in cases of early diabetic retinopathy.Methods We constructed an experimental device, which comprised an illumination optical system and a photographic optical system separated by a mirror having a hole. The device included a canon EOS 50D camera, an EF 50 mm f/1.8‐2 camera lens, a Speedlite 270EX flash, an object lens, four double‐convex lenses, three aperture stops and six artificial eyes. The eye ground was a hemisphere made of polythene terephthalate painted with six matt colour sprays: red, white brown, ochre, yellow, ivory and orange. Five fragments of house dust on the object lens and the two lenses were photographed under each artificial eye. The RGB colour space, measured by Paint Shop Pro from pictures, was changed into seven kinds of colour spaces: XYZ, CMY, HSL, HSV, HSI, L*a*b* and L*u*v*.Results The evaluation values of the following three colour spaces were favourable: the L*u*v* colour space (L*, 5.7±1.1; u*, 8.5±1.8 and v*, 6.7±1.1), the HSV colour space (hue, 2.0±0.5; saturation, 9.5±3.7 and value, 6.7±0.7) and the HSL colour space (hue, 2.0±0.5; saturation, 7.8±2.5 and lightness, 6.8±0.8). The other colour spaces did not show a good result.Conclusion The L*u*v* colour space is highly sensitive; therefore, it is most effective in distinguishing small haemorrhages from dust artefacts. The HSV and HSL colour spaces were highly sensitive in terms of saturation, lightness and value. Using Scilab and SIVP software, we are currently researching methodology that applies the use of colour spaces such as L*u*v* and HSV for automatic distinction.