A new detection operator for narrow band character extraction in low contrast scenes

Character detection plays an important role in character recognition systems. In this paper, we propose a new detection operator to extract the characters with approximately equivalent width from a low contrast simple natural scene. Initially, the Canny detector is applied to get the edge map. Afterwards, the distance transformation is performed on the edge map to get the regional maximum pixels (inner points). The distance information contributes to finding the outer contour (outer points) of the characters which compensates the gap of lacking edges. Then a new detection operator with a circular mask and a ring mask are used to act on the image along the inner and outer points to detect the characters. Our method belongs to the edge extraction category and local thresholding techniques but removes the dependency on edge accuracy. The experimental results of our tests validate the effectiveness and robustness of proposed method for various natural scenes.