Superpixel Segmentation Based on Spatially Constrained Subspace Clustering

Superpixel segmentation aims at dividing the input image into some representative regions containing pixels with similar and consistent intrinsic properties, without any prior knowledge about shape size of each superpixel. In this article, to alleviate limitation superpixel applied in practical industrial tasks that detailed boundaries are difficult be kept, we regard region independent semantic information as a subspace, correspondingly formulate subspace clustering problem preserve more content boundaries. We show simple integration conventional does not effectively work due spatial correlation within superpixel, which may lead boundary confusion error when is ignored. Consequently, devise regularization propose novel convex locality-constrained model able constrain adjacent attributes clustered generate content-aware superpixels Finally, proposed solved by an efficient alternating direction method multipliers solver. Experiments on different standard datasets demonstrate achieves superior performance both quantitatively qualitatively compared state-of-the-art methods.