The paper presents a recognition method for Sudoku Images. We use smart phone to capture Sudoku images on newspapers and books. Under different capturing conditions, Sudoku images happen several changes in gray levels and shooting angles and mix some noises. The proposed method is divided into several processing steps. We firstly use the contour characteristic of Sudoku images to extract Sudoku grid frames. Then, the images are transformed into the standard squares. Then, we partition the images into cells and extract HOG features of cells. Finally, an SVM classifier is employ to recognize digits. Because the directions of original images are variable, we add a correction procedure in the proposed method. The procedure is able to automatically rotate the images to the correct text directions. The experimental results indicate that our proposed method can quickly and recognize Sudoku images with different shooting angles and the recognition rate for digits inside cells is very high.