Aiming at the problem that text localization requires a large number oftraining samples in natural scenes

which leads to low speed of algorithmrunning and it is difficult to locate skewed text

a fast natural scene skewedtext localization algorithm based on maximally stable extremal regions (MSER)with hierarchical clustering is proposed . The method of MSER ellipse fitting is used to selectthe maximally stable extremal regions of the images

and according to thecharacteristics of the fitting ellipse and its position on the images

themajority of non-text regions are filteredout and the text candidate regions are selected. By using the idea of hierarchical clustering

thetext regions can be clustered gradually and merged into text regions rapidly.Finally the individual text regions are merged into word regions

which canachieve efficient localization of skewed scenes. Experimental results show that the speed of thisalgorithm has improved significantly without loss of locating accuracy comparedwith traditional positioning algorithms.