High-precision microscopic autofocus with a single natural image

In industrial microscopic powered operators to acquire high-quality errors in Learning-based methods: the of the prior dataset, which limits the In this paper, a high-precision autofocus distance from a single natural image. overcomes the limitations of the sharpness the dataset. Furthermore, a lightweight Defocus Prediction Model (NDPM), sufficient size was made to train all performance compared with other models, (c) 2023 Optica Publishing Group under the 1. Introduction In industrial microscopic real-time component, providing high-quality wafers [1], printed circuit boards [2], automatically, eliminating the need on analyzing the images to determine autofocus systems play an important role and focusing accuracy. One of the most common types of as shown in Fig. 1(a), which uses sharpness and determine the optimal focus position are time-consuming and sensitive to learning-based methods are proposed focusing and improve the focusing learning-based autofocus method assumes establishes a connection between the Such methods can be divided into three aspect converts autofocus into a classification stroke range and regards each sampling stack [13], two focal slices [13,14], or calculate defocus distance by the category autofocus as a regression problem. Compared the intermediate step and maps the focal general, the input is one or two focal its Fourier transform [22,23], and the position. The third aspect, which does #507757 Journal (c) 2023 Received 9 Oct 2023;