Haoran Li, Hao Zhang, Yang Zhang Stanford University, Stanford, CA

Event Info Extraction from Posters

System Pipeline

Introduction

Extraction Results

Poster Processing Steps

Future Work

Preprocess Image

Text Detection

References

An Android app that enables the user to put the event information

(time and location) into calendar by snapping a picture on the poster.

Capture  Image

Downsize  Image

Send Image to Server Preprocess  Image  (Rec7fy,  Sampling,  

Histogram  Equaliza7on…)

Text  Detec7on  (Variance,  Edge  Density,  Region  

Filtering)  

OCR  (Tesseract)  

Send String to Device NLP  Info  Extrac7on

Add  to  Calendar

Input  Image    SubSample

Edge  Detec2on    (Hough  Transform)

Find  bounding  quadrilateral  to  rec2fy

Homography  Transform

Text  Variance  &  Edge  Density

Filter  Region  (Area,  Height,  Orienta2on,  Solidity)

Bounding  Box  for  text

Input Image Rectified Image 1.   Add another function of detecting topic by

estimating font of the text in the image

2.   Use Stroke Width Transform to improve text

detection

3.   Perform better text segmentation from cluttered

background rather than simply Otsu’s Method

Text Crop OCR Result App Screenshot

1.  J. Matas, O. Chum, M. Urban, and T. Pajdla. "Robust wide baseline stereo from maximally stable extremal regions." Proc. of British Machine Vision Conference, pp384-396, 2002

2.  Tesseract OCR engine, https://code.google.com/p/tesseract-ocr/

3.  Android Calendar Tutorial, http://www.vogella.com/tutorials/AndroidCalendar/article.html