Soft and self constrained clustering for group-based labeling

When using deep neural networks in medical image classification tasks, it is mandatory to prepare a large-scale labeled set, and this often requires significant effort by experts. One strategy reduce the labeling cost group-based labeling, where samples are clustered then label attached each cluster. The efficiency of depends on purity clusters. Constrained clustering an effective way improve clusters if we can give appropriate must-links cannot-links as constraints. However, for clustering, conventional constrained methods encounter two issues. first issue that constraints not always due gap between semantic visual similarities. second attaching extra from To deal with issue, propose novel soft-constrained method, which has ability ignore inappropriate self-constrained method utilizes prior knowledge about target images set automatically. Experiments endoscopic datasets demonstrated proposed results higher purity.