Abstract:Text detection only relying on local context is not robust to separate crowd texts accurately due to the challenges exist in texts dense layouts, inaccurate annotations and complex backgrounds. T‐Skeleton explicitly exploits long‐range dependencies to address the challenges by extracting text skeleton, which is the distances transformed between text pixels and their nearest boundary pixels. T‐Skeleton has great representation capacity and distinguish long and curved texts well. Existing segmentation‐based methods have made considerable progress in arbitrarily shaped text detection due to the advantage of dealing with shape variation. However, there still exist challenges to detecting accurate text instances with dense layouts, inaccurate annotations, and complex backgrounds. Many recent works have focused on improving arbitrary boundary prediction, but it may be difficult to accurately distinguish each instance of dense layouts because their boundary pixels may be mistakenly classified to produce inaccurate results (i.e., adhesive texts) with inaccurate annotation and complex backgrounds. Considering the local and long‐range dependencies, this paper proposes an efficient text detector, namely T‐Skeleton, to obtain more reliable segmentation detections. In the spirit of object skeletonization, we introduce the text instance skeleton highlighting the semantically significant structure (similar to the skeleton of a fish) to explicitly capture the long‐range dependencies of text instances. The key idea of T‐Skeleton is to calibrate the coarse text proposals by embedding text instance skeletons to separate crowd texts accurately and robustly. We further design a channel attention module to enlarge the performance margin between T‐Skeleton and the segmentation baseline. Experimental results on four publicly available datasets show the superiority of T‐Skeleton in handling long and curved texts.

A Skeleton Pruning Method Based On Saliency Sorting

A Skeleton Pruning Algorithm Based on Information Fusion

Real-time Skeletonization for Sketch-Based Modeling

A Combinatorial Pruning Algorithm for Voronoi Skeletonization

On the Generation and Pruning of Skeletons Using Generalized Voronoi Diagrams

Learning Slimming SSD Through Pruning and Knowledge Distillation

Distilling the main skeleton line of shrimp based on machine vision technology

Loss Constrains Added Squeeze and Excitation Blocks for Pruning Deep Neural Networks

WRGPruner: A new model pruning solution for tiny salient object detection

Research on Skeletonization of Palmprint Image.

An Optimized L1-Medial Skeleton Extraction Algorithm

Skeleton Neural Networks via Low-rank Guided Filter Pruning

A Novel Method for Improving Artifacts of Chinese Calligraphy Character Skeleton Extraction

Denoising Method for Microscopic Image of Caenorhabditis Elegans Based on Skeleton-pruning

ARank: Toward specific model pruning via advantage rank for multiple salient objects detection

Structural Pruning via Latency-Saliency Knapsack

Noise-Against Skeleton Extraction Framework and Application on Hand Gesture Recognition

Fast Tree Skeleton Extraction Using Voxel Thinning Based on Tree Point Cloud

A Comprehensive Study of Structural Pruning for Vision Models

ProMask: Probability Mask for Skeleton Detection

T‐Skeleton: Accurate scene text detection via instance‐aware skeleton embedding