清晰度评价函数是自动聚焦技术的核心部分, 性能良好的聚焦曲线应该具有单峰性、无偏性、高灵敏度和抗噪性。通过将梯度差分与统计相关结合使用, 提出了一种用邻域互相关对每个像素的梯度值进行加权的算法, 并设定阈值去除贡献小的像素点。实验中使用定量指标对所提算法、一些传统算法以及一种梯度阈值算法的性能进行了评估。结果表明, 所提算法在灵敏度和抗噪性方面效果较优。

The definition evaluation function is the core part of automatic focusing technique, and the focusing curve which has better performance must be unimodality, unbiasedness, high sensitivity and anti-noise. A new function is proposed based on the combination of the gradient difference and the statistical correlation. The gradient value of each pixel is weighted by the neighbourhood cross-correlation, and the threshold is set to remove these pixels whose contributions are small. The experiment uses some quantitative indexes to evaluate the performances of the proposed algorithm、 some traditional algorithms and a gradient threshold algorithm. The result shows that the proposed algorithm has better sensitivity and anti-noise.