Study Of Text Detection Techniques Combined Local Contour Information With Learning Classification

The learning of text detection and location in image has been of great value to the description and understanding of scene contents. Therefore, an automatic tool is seriously needed for text location in image for retrieving, querying, browsing and understanding scene contents in order to increase the efficiency of scene image management. However, the background of text images is complicated and variable, which makes there is a great challenge to accurately detect the text area in image. To solve the above problem, my research work mainly focus on how to use the stroke features of text to effectively distinguish text area from background area.According to some new thoughts from home and abroad, a new text detection algorithm is proposed by analyzing the characteristics of the text, at the same time try to do a application analysis for the specific scene text detection.Firstly, analyzing the stroke directions distribution feature of text region in image, the algorithm introduces multiple scale-orient factors and the edge coarseness feature, in this paper a new text localization algorithm is proposed in a localization-to-verification framework. Mainly base on text region is usually composed of multiple oriental strokes which not only possess gradient edge but also orient distribution orderliness statistically.Secondly, taking advantage of local text style salient region descriptor and Connected-Components analysis, a robust scene text extraction approach in complex background is studyed in this paper. Salient text region is located by using local region's salient intensity variance and edge gradient in salient multiple orientations, and using SVM classifier to classificate and validate the Connected-Components of Salient Region, ultimately, based on the text Connected-Components to locate text region .Finally, studying how to use the technology of text detection for license plate locating in specific application scenes.Considering the license plate has not only the general features of the text, but also a number of inherent structural features , based on the analysis of text features, taking full integration of the color distribution features of the license plate, a robust secondary license plate location algorithm is studyed.