Similarity Criteria for Stereo-Based Surface Reconstruction in Bimodal SEM

conference paper

English

The paper is describing a robust approach to reconstruction of 3D surfaces of microscopic samples, based on bimodal stereo-image couples, that has reached a degree of maturity in its present stage of a long-term development. The method, on the difference to most of present stereo work, does not impose any a-priori constrains or models to the computed surfaces. This unconstrained approach implies using the straightforward search-and-match disparity analysis, which must be based on an extremely reliable similarity criterion. Several common criteria have been extensively tested but none has proved sufficient reliability. Therefore, the uncommon cosine similarity criterion has been introduced that represents the best so far available and probably novel solution. As learning alternatives of the criterion can be realised by means of neural networks and may thus represent a stimulation for neura-network research, the criterion analysis forms the central part of this contribution. The above mentioned choice of the criterion, together with utilising simultaneously the image information of both available imaging modalities – back-scattered electrons and secondary electrons, further implementing precautions against topological errors, and using the sc. layered structure of the disparity analysis algorithm, provides a possibility of calculating even complex surfaces with a reasonable degree of reliability.

disparity analysis, similarity criteria, image processing

JAN, J., JANOVÁ, D.

2002

1. 1. 2001

Technol. Edu. Inst. of Crete

Crete - Greece

960-85316-5-9

Proc. 4th int. conf. NNSMED 2001

1

1

275

554

280