Probabilistic homogeneity for document image segmentation

作者：

Highlights：

• Gestalt patterns are characterized in text regions to describe text homogeneity.

• A probabilistic hierarchical model is formulated to evaluate text homogeneity.

• A Bayesian cue integration model is proposed to compute homogeneity probability.

• Document segmentation using probabilistic homogeneity outperforms state-of-the-art.

摘要

•Gestalt patterns are characterized in text regions to describe text homogeneity.•A probabilistic hierarchical model is formulated to evaluate text homogeneity.•A Bayesian cue integration model is proposed to compute homogeneity probability.•Document segmentation using probabilistic homogeneity outperforms state-of-the-art.

论文关键词：Probabilistic local text homogeneity,Random walk-and-check simulation,Bayesian cue integration,Text homogeneity pattern,Document image segmentation

论文评审过程：Received 6 May 2020, Revised 6 July 2020, Accepted 9 August 2020, Available online 10 August 2020, Version of Record 17 August 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107591