In order to identify scene texts from the background interferences, many existing methods have been presented in the recent years. In this thesis, a novel text recognition method is proposed. First, Otsu edge detection is applied to the image binarization and the parameters (i.e. weights) found in a K-cluster. Second, the modified K-cluster algorithm is used to detect the text from an image. The complex background is filtered out as well. Third, the detected text gradients are evaluated by HoG (Histogram of Gradient). Accordingly, the distribution of the detected text gradients is generated. Finally, the gradient distribution is utilized by HMMs to recognize the text. The proposed approach can outperform the existing methods.