sbnair / video_text_detection

Bachelor Thesis | Text extraction from complex video scenes

Department of Engineering Informatics

Thesis Title

Text extraction from complex video scenes

Supervisor

Dr. Athanasios Nikolaidis, [email protected]

Objective

The objective of my Thesis was the development of an Image Processing desktop application, capable of detecting and extracting text displays from videos with complex backgrounds. My approach implements Machine Learning and Image Analysis methodologies from various popular scientific papers.

Technologies

JavaFX, OpenCV, LIBSVM, Gradle, JUnit, TestFX

References

[1] Palaiahnakote Shivakumara, Trung Quy Phan and Chew Lim Tan, Senior Member, IEEE “A Laplacian Approach to Multi-Oriented Text Detection in Video”, IEEE

[2] Trung Quy Phan, Palaiahnakote Shivakumara and Chew Lim Tan “A Laplacian Method for Video Text Detection”, School of Computing, National University of Singapore, 2009

[3] Rakesh Mehta, Karen Egiazarian, “Rotated Local Binary Pattern (RLBP) Rotation invariant texture descriptor”, 2nd International Conference on Pattern Recognition Applications and Methods, ICPRAM 2013, Barcelona, Spain, 2013

[4] Cong Yao, “MSRA Text Detection 500 Database (MSRA-TD500)”, Huazhong University of Science and Technology, 2012

[5] Visual Geometry Group, “Synthetic Word Dataset ”, Department of Engineering Science, University of Oxford

Download my Thesis [PDF-Greek]

Text Detection

Original Frame

Gaussian Filter

Grayscale

Laplacian Filter

Maximum Gradient Difference

Binarization

Dilation

Connected Components

1st Filter - Removing components with bigger height than width

2nd Filter - Removing components with small area

3rd Filter - Classifying text areas using Support Vector Machines

Text Extraction

Cropped Text Area

Grayscale

Unsharp Masking

Otsu Binarization

Apply OCR

JavaFX Application

Main View

Settings

Choosing video file

Text Extraction

Text Detection