Thresholding is an important process required to most tasks in computer vision especially characters’ detection from scene images. So, in this paper, we focus to enhance our previous model which is used to detect the characters from scene images and has proven superior to most existing methods. This improvement is fulfilled by replacing the thresholding method in the entire model in the stage of converting the grayscale image into the binary one. The new thresholding method is known by Bradley method. Although our previous model got better when we use Bradley thresholding, the number of false detections still less than in the case of using our previous thresholding method in the entire designed model