Font Size: a A A

Study On Text Location Algorithm In Complex Images And Its Implementation

Posted on:2016-10-14Degree:MasterType:Thesis
Country:ChinaCandidate:Y JiangFull Text:PDF
GTID:2308330461978014Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Color images often contains a lot of text information, the text information have interwovenness connection with the content of images, detecting this information is key to understand the content of images. Therefore, text detection is used extensively in Internet Filtering, License Plate Retrieval, Images Retrieval/Classification. Because color images have complex and volatile background, text’s size, font, orientation are not fixed, make it difficult to extract and recognition text with Optical Character Recognition technology. As a result, text location technology in complex images is a hot area in this field all the time. Although, forefathers have already made a good progress, the complicated and changeable of background and foreground make the problem is still not totally solved.On the basis of studying and summary previous algorithms, a new algorithm is proposed in this paper, text is located in color images precisely and comprehensively. The main work of this paper includes:(1) A new algorithm is proposed to calculate stroke width:Connected Component based Stroke Width Transformation algorithm; (2) Sparse Representation and Conditional Random Field model algorithms are implicated to text detection field, the precision of text location is improved; (3) Forward-Backward Algorithm that is improved in this paper is used to locate arbitrarily text.Connected Component based Stroke Width Transformation is an algorithm to extract a feature of letter, by analyzing character Connected Component, the algorithm is proposed. The algorithm extract text’s special Stroke Width feature quickly, this feature can be used to determine text. Sparse Representation and Conditional Random Field model are two important theory tools in pattern recognition field. In this paper, the features of Sparse Representation and Conditional Random Field model are studied, sparse reconstruct error is used to construct the parameter of Conditional Random Field model, at last, text regions are extracted precisely with Graph Cut algorithm. To locate arbitrarily text in complex scene, original Forward-Backward Algorithm is improved in this paper, text is located more roundly with improved Forward-Backward Algorithm.To confirm the effectiveness of this algorithm, experiments are simulated on ICDAR2011 database. Experiments show that this algorithm has a lot of improvement in Recall and Precision than existing algorithm. This algorithm is able to detect text in complex background roundly and precisely.
Keywords/Search Tags:Stroke Width Transformation, Sparse Representation, Conditional RandomField model, Te×t Detection
PDF Full Text Request
Related items