The goal of this project is to implement a scene text detector that can detect semantic phrases.
- Text in the scene are detexted using EAST text detector.
- Cluster words that are close by together using K-Means Clustering Algorithm
- Sort words the order of words similar to human reading conversions for English.
- Need to know the number of clusters beforehand.
- Low performance bounding box detection of text
- Lack of Layout context like fonts and font size.
- Use cluster-agnostic clustering algorithm
- Use more accurate text detectors
- Integrate background context, Font and Layout information