The framework of the thesis Indexing using Text

The framework of the thesis

Indexing using Text

Plan

Videos vs. scanned documents

What is text? - character segmentation

What is text? - texture

What is text? - contrast & geometry

A text detection system for videos

Plan

2 Algorithms for still images

The local contrast method

Properties of the local contrast method

Method 2: why learning?

Geometrical features

Geometrical features: baseline

Estimation of the text rectangle height

Features

Learning with Support Vector Machines

Plan

Tracking the text appearances

Tracking: content verification

Enhancement

Plan

Adaptive binarization

Our solution: contrast maximization

Character segmentation: examples

Modeling text with a Markov random field

The prior knowledge

Plan

Evaluation measures

Detection in still images

The influence of falling generality

Detection in video sequences

OCR results

TREC 2002

Conclusion

Outlook