In the hot continuous casting billet surface defect inspection system based on machine vision, acquiring a high signal-to-noise image is the key for successful inspection. To solve the disadvantages existing in current machine vision engineering, a new algorithm with improved image definition is presented based on both focus window and CCD target area illumination parameters. It selects a target object from series of hot continuous casting billet surface images, and then acquires the optimum articulation through focus window square gradient algorithm. By recognizing and calculating target’s area loss rate, target area parameter evaluation can be done. The global optimum image quality point is achieved. The algorithm is effective in selecting focus plane and shutter time during hot continuous casting surface imaging process and is of a good practical value. At the same time, the algorithm is useful for image collecting work in other machine vision engineering.