A PC-Based Interactive Image Processing System for 3-D Analysis of SEM Stereo Images

A PC-based interactive image processing system for 3-D analyses of stereo images of a scanning electron microscope (SEM) using an advanced feature-based stereo model is described. Integration between the global and qualitative human vision and the local and quantitative computer vision makes it possible to greatly reduce the computational load and to highly improve the reliability of stereo correspondence. The system consists of a P C with a co-processor and a frame buffer, an image scanner. a mouse. and an analogue color CRT. The stereo correspondence algorithm, implemented on the PC with MS-C.is carried out from point to point interactively referring to a stereoscopic display to obtain 3-D digitized data. Ceometrically defined properties including surface orientations, sectional profiles, depths and thicknesses are analyzed based on the 3-D data of necessary points. Some experimental results are presented for crystal analysis. surface 3-D shape recognition and thickness measurement. Discussions are also presented on accuracies in 3-D measurement. Method (1) is suitable for surfaces with smooth curvatures. but not applicable to common surfaces with edges or steps where the gradient becomes indefinite or infinite. such as edges of crystals and steps of VLSl patterns. In addi tion. the method requires a specially equipped SEM. Method (2) can be carried out with a common SEM, but it requires so much and so complex computation in stereo correspondence that introduction of a large-scaled image processor is necessary and, nevertheless. considerable errors are inevitable in order to reconstruct 3-D shapes over the whole image. The state-of-the-art situation is that neither method can respond to the demands to obtain 3-D information from SEM images. Surveys of the demands have revealed that the demands include not only 3-D reconstruction on the whole image plane, but also 3-D measurements of geometrical properties solely on particular areas to be investigated. This paper describes the introduction of an advanced feature-based stereo model and interactive processing from point to point to the method (2) which makes it possible to obtain stereo correspondence and to reconstruct 3-D properties in real time even on PC.