Robust video text segmentation and recognition with multiple hypotheses