-
Notifications
You must be signed in to change notification settings - Fork 0
GSoC 2014 : Sandeep Kaur Text Detection Algorithms
Sub-organization with whom you hope to work for: scikit-image
Name : Sandeep Kaur
Email : mkaurkhalsa@gmail.com
Telephone : +91 959 204 0370
Timezone : India (UTC+05:30)
Source control user name : sandeepmadaan [GitHub]
IM Information
Skype : sandymadaan
Twitter : sandymadaan
Blog
Blog URL : http://sandymadaan.wordpress.com/
University : Punjab Technical University, Punjab(India)
College : Shaheed Bhagat Singh State Technical Campus, Ferozepur
Major : Computer Science
Current Year : First Year
Expected Graduation : 2015
Degree : M.Tech (Masters in Technology)
scikit-image: Text Detection Algorithm
Detecting text in natural images is an important prerequisite. Out of many methods for text detection, we employed edge-enhanced Maximally Stable External Regions as basic letter candidates. These candidates are then filtered using geometric and stroke width information to exclude non-text objects. A noval image operator seeks to find the value of stroke width for each image pixel. Letters are paired to identify text lines, which are subsequently separated into words.
To solve our purpose, we use. :
Edge - Enhanced MSER algorithm
and Stroke Width Transform method
The various milestones for this job are described as :
March 21 - April 21 ( Interim Period )
- Learn more about scikit -image tools and its features. Get to know how whereas algorithms in it were implemented.
- Remove bugs
- Implement new features
- Interact more on mailing list.
21 April - 18 May ( Community Bonding Period )
- Discuss the algorithms to be implemented with mentors
- Learn more about stroke width and MSER edge detection.
19th May - 1 June
- Implement Connected Component(CC)-based text detection algorithm, which employs Maximally Stable External Regions(MSER). The implementation of MSER can be done using this.
- Write the test cases.
2 June - 10 June
- Despite the favorable properties, MSER have been reported to be sensitive to image blur. To allow for detecting small letters in images of limited resolution, the complimentary properties of Canny edges and MSER are combined in our edge-enhanced MSER. Thus the complimentary properties of cany edged and MSER are calculated and combined.
11 June - 23 June
- We have now got foreground CC's which are considered as letter candidates. We perform a set of simple and flexible geometric checks on each CCs to filter out non-text objects.
- Now we develop the image operator to calculate the stroke width.
24 June - 14 July
- Using the SWT operator developed above the text width is found.
- Thus the text lines and words are separated out.
- Write test cases.
15 July - 31 July
- The stroke width method done above does not work well when the opposite stroke edges are not parallel in those cases the Euclidean distance transform is applied to label each foreground pixel with the distance to its nearest background pixel. Implementation of this algorithm is done.
- Write test cases.
- After find the stroke width, the text line aggregation and words are detected.
1 August - 14 August
- Code clean up.
- Documentation.
15 August - 18 August
- More testing
- Check errors in Documentation
19th August - 26 August
- Pencil Down and final Evaluation
- I have been collecting user requirements for long. I also have read many research papers, found alternatives of text detection and selected best amongst them.
- I have read and understood the code of scikit-image. Also read its examples which will be usefull in text detection algorithm.
Because I am good at Python and I love Image Processing. Earlier I used to do Image processing with Gimp, but combining my love and skill will be awesome.
- I am open to Ideas, I don't stick with tools but tasks.
- I have read and understood most of scikit-image code and will be adding new features into it(I have also been studying their algos)
- I will be doing tasks given to me even befor and after the GSoC.