Research On Text Location Agorithm In Complex Background Image

At present, a method how to quickly and accurately recognize character from a image with complex background is still a hotspots that many scholars from home and abroad concen. Due to a large extent, the system recognition performance often depends on the accuracy of the image text region location, so the text location is a critical step for text identification, which directly affect the identification results of system. In view of different text location targets and different outstanding issues, moreover, the existing text location algorithm lack a general standard database and evaluation criterion, so massive intensive study should been done by scholars to get a text location algorithm with a high detecting probability, low undetected rate and false detect rate.Combined with the existing image text location technology, a morphological text location method based on edge and filtering is presented in this paper. Use the knowledge of edge and filtering to extract initial text block and enlarge the initial text block to get the candidate text area, then use the vertical Sobel operator to extract stroke edge and the edge density is connected with stroke edge to extract rectangle edge, in addition, introduce the vertical and horizontal projection to accurately locate the initial text area, at last, according to morphological limit and connected domain to accurately get the text region. Experimental results show that this method has high positioning accuracy and extraction rate, moreover, it has a accurate text location region and rapid positioning speed.Meanwhile, in view of the features that corner information of text region characters is abundance, dense and relatively in order, this paper also proposed another text location algorithm based on adaptive Harris corner detection. Since detected corner mostly is pixels and the corners appeared in text region is more than non-text region, a significant characteristic that the occupancy how many pixels in a certain region area is brought in this paper to judge whether the got connected region is text region. Experimental results show that this algorithm can avoid the choice of threshold, effectively overcome corner redundance and missing caused by wrong threshold chose, and improve accuracy of corner detection, thus achieve the quick accurate location of follow-up text region.