Text localization using DWT fusion algorithm

Abstract

Texts provide highly condensed information about the contents of images or video sequences. It proposes a text localization method using discrete wavelet transform and neural network. Because of the property of Harr wavelet transform, the proposed approach can localize large font size text by merging 4 small block wavelet coefficients into large one. Then the morphological dilation operation and the neural network are employed to raise the recall rate and precision rate. Experimental results show that the proposed method can successfully locate text region from complex images. Its precision rate is better than that of other methods.