Figure 3.

Some meibographic images taken from both upper and lower eyelids and analysed with the four automatic and manual detection methods. The green region for manual analysis shows glands region and red represent loss area, while in case of automatic analysis methods white region represents gland area and colour regions represent loss area. Results revealed that the automatic detection method percentage results are almost on par with the manual analysis. In manual analyses, the analyser while putting dots or lines around the glands is more likely to skip some minor regions between the glands while the automatic analysis considers this region in the analysis. The cause of this difference is due to scare tissues and light reflection on the images and system classify them as meibomian glands area. We can see from figure 3 that MG-GAN outperformed state of the art detection methods for MG detection. CGAN, conditional adversarial neural network; GAN, generative adversarialnetwork; MG-GAN, meibomian gland-generativeadversarial network.