Shape from focus is a commonly-used method for measuring the three-dimensional topography of components with small size. However, the measurement accuracy is often affected by noise in the focus evaluation curve. Additionally, the efficiency is limited by both the number of image sequences and the focus evaluation algorithm. In this paper, an adaptive weighted focus extreme search method combining the theory of dual-fitting optimization and a focus unimodality evaluation index is proposed. Furthermore, a cubic polynomial quadratic interpolation algorithm is used for interpolation, and the depth information is optimized by a multi-level median blend filtering algorithm. Experimental results show that the proposed method reduces the RMSE of simulated image data by 19. 08% and 17. 32% compared to Gaussian fitting and polynomial fitting, and maintains three-dimensional resolution in regions with significant noise interference. At the same time, we also measured the drill bits. Results show that the proposed method reduces the RMSE by 77. 8% and 62. 6% even when the number of image sequences and image resolution are reduced by 50% , which effectively reduces the influence of measurement curve noise on measurement results and improves measurement efficiency.