Image binarization is one of the main techniques for image segmentation. It segments an image into foreground and background. The foreground contains interested objects. Usually, the binarization is carried out with a threshold found from the histogram of an image automatically. It has many applications in pattern recognition, computer vision, and image and video understanding. This paper formulates the binarization as an optimization problem: finding the best threshold that minimizes a weighted sum-of-squared-error function. A fast iterative optimization algorithm is given to reach this goal. Our algorithm is also compared with a classic commonly-used binarization method. The experiments show that the two algorithms yield the same segmentation results but our algorithm is more efficient.