针对目前调焦评价函数存在的灵敏度低、鲁棒性差、易进入饱和区等问题, 提出了一种融合灰度梯度和暗通道L0范数的调焦评价函数。首先分析并证明了清晰图像与离焦图像相比具有亮度更大、更稀疏的暗通道这一特性, 然后选择了暗通道的L0范数对图像的离焦情况进行评价, 最后融合图像灰度梯度与暗通道的L0范数构建了一种新的调焦评价函数。在对森林、公路、天空中的无人机、室内静物等多种场景的调焦实验中, 使用单核DSP处理速度为9 Hz, 基本满足实时性; 受到外界干扰时函数曲线没有出现伪峰; 在离焦比较严重的情况下没有进入饱和区, 仍然能够保持一定的陡峭度。实验结果表明, 本文提出的调焦评价函数具有较好的灵敏度和鲁棒性, 可以满足光电成像系统对实时调焦的要求。

In order to solve the problem of current focus measure function, such as low sensitivity, poor robustness and easy to enter the saturated zone, a new function based on gray gradient and dark channel is proposed. Firstly, a characteristic is analyzed and proved that a clear image has brighter and sparser dark channel compared with a defocus image. Secondly, the L0norm of dark channel is selected to estimate the degree of defocus. Finally, a new focus measure function is constructed by integrating the gray gradient and the L0norm of dark channel. The processing speed of single core of DSP is 9 Hz, which basically meets real-time performance in focusing experiments on various scenes such as forests, roads, drones in the sky, indoor still life, etc. Also, pseudo peak cannot be observed when the focusing process is disturbed and the function can still maintain a certain degree of steepness when the defocus is serious. A series of experiment results showed that the focus measure function proposed in this paper has good sensitivity and robustness, which can meet the real-time focusing requirement in electro-optical imaging system.