Detecting text from natural images with Stroke Width Transform

18
Detecting Text in Natural Scenes with Stroke Width Transform Presented by, POOJA G N

Transcript of Detecting text from natural images with Stroke Width Transform

Page 1: Detecting text from natural images with Stroke Width Transform

Detecting Text in Natural Scenes with Stroke Width Transform

Presented by,

POOJA G N

Page 2: Detecting text from natural images with Stroke Width Transform

Overview• Introduction• Steps involved in text detection algorithm• Edge map• Stroke width transform• Finding letter candidates• Grouping letter candidates• Strength and weakness of SWT• Results• Applications• References

Page 3: Detecting text from natural images with Stroke Width Transform

Introduction• With the increasing use of digital image capturing devices,

content-based image analysis techniques are receiving intensiveattention in recent years.

• As indicative marks in natural scene images, text informationprovides brief and significant clues for many image-basedapplications.

• We present a image operator that seeks to find the value ofstroke width for each image pixel, and demonstrate its use onthe task of text detection in natural images.

Page 4: Detecting text from natural images with Stroke Width Transform

Introduction(contd.,)Current text detection approaches can be roughly classified into three groups:

Region-based approachesThis attempt to use similarity criterions of text, such as color, size, strokewidth, edge and gradient information, to gather pixels.

Texture based approachesThis utilize distinct textural properties of text regions to extract candidatesub-windows and the final outputs are formed by merging these sub-windows.

Hybrid approachesThis take advantages of both region-based approaches which can closelycover text regions and texture-based approaches which can estimatecoarse text location in scenes.

Page 5: Detecting text from natural images with Stroke Width Transform

Steps involved in text detection algorithm

Page 6: Detecting text from natural images with Stroke Width Transform

1. Image(input)

Page 7: Detecting text from natural images with Stroke Width Transform

2. Edge map Here we use Canny Edge detection algorithm. The Canny edge detector is an edge detection operator that uses

a multi-stage algorithm to detect a wide range of edges in images.

Input image Edge detected image

Page 8: Detecting text from natural images with Stroke Width Transform

3. Stroke Width TransformSWT is a local operator which calculates for each pixel the width of the most likelystroke containing the pixel.

(a).

(b).

(c).

Figures shows the implementation of the SWTwhere(a) A typical stroke. The pixels of the stroke in

this example are darker than the backgroundpixels.

(b) p is a pixel on the boundary of the stroke.Searching in the direction of the gradient atp, leads to finding q, and thecorresponding pixel on the other side of thestroke.

(c) Each pixel along the ray is assigned by theminimum of its current value and thefound width of the stroke.

Page 9: Detecting text from natural images with Stroke Width Transform

The rules to components are as follows:

• The variance of the stroke-width within acomponent must not be too big.

• The aspect ratio of a component must be within asmall range of values, in order to reject long andnarrow components.

• Components whose size is too large or too smallwill also be ignored.

4. Finding Letter Candidate

Page 10: Detecting text from natural images with Stroke Width Transform

5. Grouping letter candidates into regions of text

• Grouping the pixels into letter candidates based on their stroke width.

• The grouping of the image will be done by using a Connected Component algorithm.

• The image partition creates a set of connected components from an inputimage, including both text characters and unwanted noises.

• We perform structural analysis of text strings to distinguish connectedcomponents representing text characters from those representing noises.

• Assuming that a text string has at least three characters in alignment, wedevelop two methods to locate regions containing text strings: adjacentcharacter grouping and text line grouping.

Page 11: Detecting text from natural images with Stroke Width Transform

Grouping letter candidates into regions of text(contd.,)• Group closely positioned letter candidates into regions of text.• Filters out many falsely-identified letter candidates, and improves the

reliability of the algorithm results.

The rules to pair the letters are as follows:

• Two letter candidates should have similarstroke width.

• The distance between letters must notexceed three times the width of the widerone.

• Characters of the same word are expectedto have a similar color; therefore wecompare the average color of the candidatesfor pairing.

Page 12: Detecting text from natural images with Stroke Width Transform

Resultant Image at each step of the algorithm

Page 13: Detecting text from natural images with Stroke Width Transform

Strengths of SWT• The SW Detector can detect letters of different languages (English, Hebrew, Arabic etc.)• The text can be of varying sizes.• The text can be of different orientation, including curvy text.• Even handwriting can be detected.

Weakness of SWT• Appearance of noise.• Foliage resembles letters.• Does not handle round and curved letters.• Small and close letters tend to be grouped together in the SW labeling phase and these

groups may be dismissed in the ‘finding letter candidates’ phase.

Page 14: Detecting text from natural images with Stroke Width Transform

Results

Page 15: Detecting text from natural images with Stroke Width Transform

Applications

Mobile text recognition Content-based web image search Automatic geocoding Robotic navigation License plate reading

Page 16: Detecting text from natural images with Stroke Width Transform

References1) Gili Werner ”Text Detection in Natural Scene with Stroke Width Transform”. ICBV,

February, 2013.

2) B. Epshtein, E. Ofek, and Y. Wexler, “Detecting text in natural scenes with strokewidth transform,” in Computer Vision and Pattern Recognition(CVPR),Conferenceon. IEEE, 2010.

3) Mr. Hemil A. Patel, Mrs. Kishori S. Shekokar, “Text Detection in Natural Scenes withStroke Width Transform”, [Patel, 3(11): November, 2014], ISSN: 2277-9655.

4) L. Neumann, J. Matas, “ A method for text localization and recognition in real-worldimages”, ACCV, 2010.

Page 17: Detecting text from natural images with Stroke Width Transform

Any queries?

Page 18: Detecting text from natural images with Stroke Width Transform

Thank you