常规的尺度不变特征变换（SIFT)图像特征提取方法难以提取多聚焦图像离焦模糊区域的特征，使得图像间存在局部、少量的公共特征，导致多聚焦图像配准精度差，严重影响后续图像融合和三维重建质量。在分析图像离焦模糊区域特征提取不确定性的基础上，提出了一种多聚焦图像离焦模糊区域的SIFT特征提取方法。首先提取多聚焦图像聚焦清晰区域的SIFT特征，再利用光流跟踪提取对应离焦模糊区域的SIFT特征，避免了在离焦模糊区域直接提取SIFT特征的不确定性。实验结果表明：提出的方法在离焦模糊区域具有良好的SIFT特征提取能力和提取精度，能实现多聚焦图像SIFT特征匹配数量显著增长，SIFT特征提取的误差为0.03~0.39 pixels，优于现有方法的0.21~1.71 pixels。降低了离焦模糊区域SIFT特征提取的不确定性，为多聚焦图像精确配准奠定了基础。

Overview: Two-photon lithography (TPL) has been a research hotspot in 3D micro/nano writing technology due to its characteristics of high resolution, low thermal influence, a wide range of processed materials, low environmental requirements, and 3D processing capability. It has shown unique advantages in the fields of life science, material engineering, micro/nano optics, microfluidic, micro machinery, and so on. This paper summarizes the research works done by researchers on different writing methods to improve TPL processing efficiency. Single-beam writing is the main method for TPL, which mainly depends on the speed of the scanning device. Single-beam writing has the advantages of simple system and high-quality beam, and it is easy to combine various effects to improve writing results. It mainly includes scanning modes based on the translation stage, galvo, polygon laser scanner, and acousto-optic deflector (AOD) (Fig. 2). All these modes have advantages and disadvantages. As for the scanning speed comparison, polygon laser scanner and AOD have relatively faster writing rates (faster than m/s). Multi-foci parallel lithography can obviously promote efficiency, elevating the speed by dozens or even hundreds of thousands of times, mainly based on spatial light modulator (SLM), digital micromirror device (DMD), microlens array (MLA), diffractive optical elements (DOE), multi-beam interference, and so on (Figs. 3-15). Multi-foci parallel lithography based on SLM is most widely used owing to its high efficiency and ability to flexible and independent control of each single beam, but the refresh rate is still insufficient. DMD has a higher refreshing rate (32 kHz), but the state-of-the-art beam parallelism realized by DMD is severely limited. More parallel beams are further required for improving the processing efficiency. The 2D pattern exposure method based on SLM or DMD can further improve the TPL efficiency with the superiority of generating flexibly designed pattern (Figs. 16-18). However, the 2D projection exposure technology is still difficult to achieve high writing precision, especially the axial resolution. An available method to improve the axial precision is spatially and temporally focusing an ultrafast laser to implement a strong intensity gradient at the spatial focal plane that restricts polymerization within a thin layer. The 3D projection method will be the most efficient writing method in the future, especially in 3D device processing (Figs. 19-20). Researchers used this technique to make hollow tubular and conical helices structures, increasing the processing speed by 600 times. However, the research results show that the current 3D projection can only process simple 3D structures. Further researches on 3D exposure processing of complex structures are expected, which will effectively expand its application in various fields. Authors believe that with the effort of researchers on efficiency improvement gradually, TPL can further highlight its advantages to promote the development of life science, materials engineering, micro-nano optics, and many other fields.

针对传统多聚焦图像融合算法中融合边缘出现模糊、伪影等问题，提出了一种结合改进拉普拉斯能量和（SML）与差分图像的多聚焦图像融合算法。首先，为了提取源图像的聚焦特征信息，分别通过SML和滤波差分进行聚焦度量，再采用引导滤波获得更多的细节特征；接着，利用像素最大值规则生成初始融合决策图，再对初始融合决策图进行小区域去除消除因聚焦和散焦区域相似造成的噪点，并对融合决策图进行不一致处理，获得更精确的聚焦区域；最后，由逐像素加权平均规则，得到融合图像。实验结果表明，所提出的算法在主观视觉效果和客观评价指标上均优于对比算法，互信息、特征互信息、图像梯度特征在彩色图像上分别提高了0.17%、0.38%和0.11%，在灰度图像上分别提高了0.7%、0.69%和0.33%，并且平均运行时间少于0.5 s，具有较高的计算效率。此外，该算法能够较好地保留源图像信息的完整性，融合图像边缘清晰、无伪影。