Font Size: a A A

Study On The Text Detection Based On Fuzzy Classification For The Complexity Of The Image Background

Posted on:2016-01-19Degree:MasterType:Thesis
Country:ChinaCandidate:M S ChenFull Text:PDF
GTID:2298330467457357Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
In recent years, the development of the multimedia and the popularityof smart mobile devices is made digital image and digital video been foundeverywhere, and the corresponding digital processing technology also hasmade great progress. The life of people is more and more inseparable fromthe digital media, and at the same time the technology innovation to dealwith digital media t can also help people relax and live happily.Content words are essential in life which is also an important partof digital image and digital video. Positioning text content from thedigital image or digital video can help people understand the contentsof the images or video quickly and help can speed up image or videoretrieval. The text location and recognition by mobile devices can helpreading and can also be used in intelligent transportation, touristguiding purposes. This paper mainly studies the text detection andpositioning with the natural scene video. The main research contentsinclude building fuzzy classifier based on background complexity anddesigning three kinds of text localization algorithms based on backgroundcomplexity.The background of the natural scene text image is protean. For avariety of natural scene text images of complex background singlealgorithm is difficult to be applicable to all complex backgrounds, sothis paper puts forward a method of building fuzzy classifier based onthe complexity of background of the natural scene text images. By trainingthe natural scene text images the images can be divided into simplebackground images, moderately complex background images and complexbackground images according to the complexity of the background. Textdetection classifier detects the images according to the complexity ofthe background and distributes the class labels to mark the text image categories.For the simple background natural scene text images stroke edgedetection and morphology algorithm are used to position text area. Becauseof the background is very simple, the text is more noticeable in the images.The text area can be located by stroke edge detection and morphologicalprocessing. For moderately complex background of nature scene text imagethe pyramid decomposition algorithm of edge detection and morphologyalgorithm are used. The pyramid decomposition algorithm overcomes theinterferences of characters in different sizes slightly in moderatelycomplex background and the complex object which belongs to background andthen the text areas are located. For complex background text image usingstrokes width transform and connected domain analysis algorithm are used.According to the text characters stroke widths are similar to each otherthe stroke width transform algorithm is used to create the stroke widthimage. Connected domain analysis algorithm and the rules are used tofilter out noise, and then the text areas can be located finally. Accordingto the classification results adaptive the text location algorithm areadaptively adopted, which not only guarantees the locating accuracy ofthe text, but also improves the efficiency of text detection.
Keywords/Search Tags:Text detection, fuzzy classifier, Stroke edge detection, Stroke widthtransformation
PDF Full Text Request
Related items