The Research On An Adaptive LBP Based Algorithm For Image Text Detection

Fast and accurate localization of text blocks is an essential step for the extractionof Character information in images. The text which may be in various sizes, styles andalignments is as complicated as background, which is the main challenge in textdetection technique. At present, there is too much research to do. But the feasibility,accuracy and real-time of these methods can not be satisfied simultaneously. Researchon fast and robust algorithm makes sence. In this paper, we focus on studying efficientalgorithm for video text detection while with low complexity. The principal works areas following:(1) For the shortcomings of poor adaptability of the existing LBP operator, weproposed an adaptive LBP threshold algorithm. The algorithm uses the divideddifference of the values between global and local grey pixels to determine the size ofadaptive threshold so that it can maximize the removal of background. Experimentalresults demonstrate that the proposed approach has the strict monotone invariance andstrong texture classification ability, while with low complexity.(2) For the shortcomings of high complexity of the existing region growingalgorithm, we proposed the region growing algorithm with low complexity. To reducethe complexity, the method uses approximate round masking to cover multi-directionjudgments for obtaining the seed pixel. Then we combine the adjacent pixel to formthe text region. Experimental results demonstrate that the modified region growingalgorithm is equivalent to the existing region growing algorithm, while it has lowcomplexity.(3) In order to verify effectiveness and feasibility of the proposed algorithm, thefour methods, which are the existing LBP operator, adaptive LBP threshold algorithm,the existing region growing algorithm and the region growing algorithm with lowcomplexity, are combined to make experimental analysis with the correspondingperformance statistics. Experimental results demonstrate that the proposed approachcan detect characters accurately even in video with complex background, and it is ofstrong adaptability and practicality, while with low complexity.