Survey On Text Detection Under Complex Background

Text in the image plays a very important role in applications such as semantics understanding and image retrieval. It also has a wide development prospect in applications such as the image retrieval under Internet and the discernment of the number plate in traffic administration. So text location is the basis of the application which is mentioned above. To obtain the text, text location method can be used. Currently, the text location technology is not mature enough to meet the requirement of the practical application. Therefore, the objective of this dissertation is to solve the problem of text location under complex background.This dissertation discusses the previous related work at first, then expound the text location methods detailed in three parts: text location based on the feature extraction; text location based on machine learning; text location based on the combination of the feature extraction and the machine learning. After that, a brief analyze for the advantage and disadvantage of the three parts is made. Take the complicated background image as research objects, this dissertation comes up with three methods for text location:(1)In order to solve the problem, a text location method based on active contour model is proposed. Firstly, the input image is shaped through sobel-laplacian, and then filtered by gaussian-laplacian. Secondly, the image which has pretreatment obtains the initial contour after using the improved active contour model, which can expand or narrow the contour to get the final contour by iterative algorithm. Finally, the after-treatment operation is applied to exclude non-text block and to get the final text area.(2) In order to solve the problem, a text location method based on saliency detection is proposed. Firstly, the method detects the image background by using the nodes as the benchmark which extract from the four image edges respectively. Secondly, the nodes of the foreground regions are used as the benchmark to get the final text area. Finally, the center segmentation algorithm is combined with the final text area to obtain the final text.(3) In order to solve the problem, a text location method based on edge detection with Embedded Confidence is proposed. Firstly, the method introduces confidence measure and uses it for the edge detection which is based on gradient. Secondly, nonmaxima suppression and hysteresis thresholding can be used to obtain the final image text.Text location has been a hotspot which has broad application prospect. Currently, many location methods are not ambition in complicated background and changeful text. The results of the practical application show that proposed methods in this dissertation are effective, so they have import theoretical value and practical value for future development in text location domain.