从光学系统出发介绍了基于图像处理的离焦图像成像原理,分析了传统基于灰度梯度的自动调焦算法的缺陷——灰度梯度考虑不全面并且易受噪音影响.针对传统算法不足，提出了基于图像灰度梯度最大值累加的自动调焦算法.该算法通过计算中心像素的八邻域灰度梯度，以自动聚焦区域内各像素最大灰度梯度与最大灰度与最小梯度差值乘积之和作为新的评价函数，分析了该算法可以抑制噪音影响的原理；介绍了焦点搜索自适应变步长爬山算法原理和实现过程；最后进行算法性能实验验证.首先，利用不同算法对序列图像和加入高斯噪音的图像进行清晰度评价，结果证明该算法对含噪图像的清晰度评价准确，与传统算法相比该算法在抑制噪音方面具有优越性；其次，用不同聚焦算法进行序列图像自动调焦实验，结果显示传统算法会出现错误聚焦的情况，而该算法聚焦准确迅速，相比于传统算法具有明显优势.

The principle of defocused imaging was introduced based on optical system. The defects of classical auto focusing algorithms based on gray gradient were presented, which were not considering all the gray gradient and affected by noise easily. According to the defects of classical algorithms, a new auto focusing algorithm based on largest gray gradient summation was proposed, which was realized through adding the product of the maximum and the difference of the maximum and the minimum of the eight neighborhood gray gradient of each pixel in the selected auto focusing region. The principle of restraining noise of the new algorithm was discussed. The principle and progress of adaptive variable step climbing algorithm for focusing position searching were introduced. At last, experiments were carried out to test the performance of the new algorithm. Firstly, by calculating the definition value of the original and Gaussian noise added sequences images using new algorithm and classical algorithms, and comparing the results, the advantage of restraining noise of the new algorithm was proved, for using the new algorithm, the definition evaluation of noisy image was calculated better compared to all the other algorithms. Secondly, experiments of auto focusing using different algorithms were carried out, and the results showed that using classical algorithms might focus wrong, while the new algorithm could focus correct and rapidly, which was an obvious advantage comparing to classical algorithms.