Optical profilometry techniques find a wide range of industrial and scientific applications in many fields owing to their non-contact characteristics and high-speed measurements. It is an important method in reconstructing sample surface profiles requiring high quality. Many optical profilometries have been developed for surface reconstruction in various applications. When these techniques are used to measure spectacular samples, a maximum measurable surface gradient established by the numerical aperture of the optical system must be considered. When the limit is exceeded, the optical system can only collect the scattered light from the sample but not the direct reflected light. This may deteriorate the surface reconstruction quality by reducing the signal-to-noise ratio (SNR). Besides the optical aberration of the microscope, there exist other incidents like surface tilting that may cause undesired measurement errors even if the measurement is conducted within limits.
Various methods have been developed to overcome these problems. For example, advanced optical models have been introduced to enhance 3D imaging on complex surfaces by defining the physical process involved. Theoretical and experimental breakthroughs involving 3D transfer functions and thin foil models have also been demonstrated. Other methods, such as the aperture-coded confocal microscope, have been proposed to resolve the surface slope and curvature–induced errors. However, these and many other methods suffer from numerous drawbacks, most of which are associated with undue measurement errors caused by the formation of the local surface gradient by the measured points and adjacent surface geometries.
To address these problems, Dr. Guo-Wei Wu and Professor Liang-Chia Chen from National Taiwan University proposed a single-exposure microscopic profilometry based on diffractive image microscopy (DIM) for precise 3D profile reconstruction of surface geometries. Most importantly, an artificial neural network (ANN) approach was adopted to develop the reverse mapping model by training the collected diffraction images and learning the complex mapping relationships between the images and their corresponding surface orientations. The model was then used to transform the diffraction measurements into surface geometries like tilting direction, tilting angle and height. The work is currently published in the research journal, Optics and Lasers in Engineering.
The authors showed that the proposed new method could simultaneously measure the tilting direction, tilting angle and height of the local surface with enhanced precision. Unlike other methods, the present one does not require prior knowledge of the information about the neighboring geometries adjacent to the measured point during 3D reconstructions. In addition, it achieved a surface inclined angle and height repeatability of 0.037 °C and 0.257 µm, respectively, suggesting the realization of precise 3D microscopic profilometry. Furthermore, by using the spatial light modulators to parallelize the single exposure measurement, full-field surface profilometry was achieved.
According to the authors: The significance of the new approach is in truly 3D surface reconstruction for simultaneously detecting surface position and its orientation. In contrast, traditional optical methods measure the surface depth (Z) with its spatial position (X, Y) and infer surface orientation from its measured (X, Y, Z). However, the main limitation with these kind strategies is that the measured depth may be affected by the surface orientation associated with the measured surface point and it could be erroneous. Without resolving the surface orientation simultaneously with the 3D measurement of the point underlying detection, it would be difficult to resolve the above recursive problem. In summary, Dr. Guo-Wei Wu and Professor Liang-Chia Chen developed a novel profilometry for accurate measurement of the 3D microscopic surface geometries in a single exposure manner. The feasibility of extracting the vital characteristics of the diffraction images and determining the surface orientations using ANN at the sub-micrometer level with high measurement accuracy and precision was experimentally verified on real objects. In a statement to Advances in Engineering, the authors explained that the precision of the optimized model could be easily optimized based on the application needs, making the proposed model a promising candidate for precise optical profilometry in many fields.
Wu, G., & Chen, L. (2021). Precise 3-D microscopic profilometry using diffractive image microscopy and artificial neural network in single-exposure manner. Optics and Lasers in Engineering, 147, 106732.