Abstract: To detect text from images with different backgrounds,a new algorithm based on saliency detection via graph-based manifold ranking and central segmentation was proposed. The image elements (pixels or regions) of input image are ranked by similarity with foreground cues or background cues at first.The boundary prior is exploited by using the nodes on each side of image as labelled background queries,and then binary segmentation is applied on the achieved saliency map and the labelled foreground nodes are taken as salient queries.The accurate alternative text area can be obtained at last.Then the central segmentation algorithm is used to obtain the precise edge graph.As the saliency map cannot locate the boundary and the edge graph cannot locate the text areas,we integrated both of them and got the final result.Experiment shows that the proposed method can effectively detect the text in the image.

Key words: Text detection,Saliency detection,Central segmentation