The interim connection space based on human color vision for spectral color reproduction.

To improve the spectral image color reproduction accuracy, two novel interim connection spaces (ICSs) were proposed. The dominant structure of spectral power distributions was extracted by principal component analysis for the widely used illuminants and light sources, and then further transformed to three synthetic illuminants. The CIEXYZ tristimulus under two or three synthetic illuminants was employed to construct two novel ICSs. The two ICSs were compared with LabPQR and the ICS with two sets of tristimulus under two real light sources according to the spectral and colorimetric representing accuracy of Munsell and Natural Color System (NCS) chips. The results indicated that the two ICSs proposed in this study outperformed the other two ICSs as a whole.

Pseudo-labeling generative adversarial networks for medical image classification.

Semi-supervised learning has become a popular technology in recent years. In this paper, we propose a novel semi-supervised medical image classification algorithm, called Pseudo-Labeling Generative Adversarial Networks (PLGAN), which only uses a small number of real images with few labels to generate fake images or mask images to enlarge the sample size of the labeled training set. First, we combine MixMatch to generate pseudo labels for the fake and unlabeled images to do the classification. Second, contrastive learning and self-attention mechanisms are introduced into PLGAN to exclude the influence of unimportant details. Third, the problem of mode collapse in contrastive learning is well addressed by cyclic consistency loss. Finally, we design global and local classifiers to complement each other with the key information needed for classification. The experimental results on four medical image datasets show that PLGAN can obtain relatively high learning performance by using few labeled and unlabeled data. For example, the classification accuracy of PLGAN is 11% higher than that of MixMatch with 100 labeled images and 1000 unlabeled images on the OCT dataset. In addition, we also conduct other experiments to verify the effectiveness of our algorithm.