Cross-scale focus measure aggregation in depth recovery of microscopic objects

In shape from focus (SFF) methods, image focus volume enhancement is an important issue for acquiring accurate depth maps. Mostly, conventional approaches do focus aggregation locally to enhance the focus volume, which may not suppress noisy focus measurements properly and consequently, may provide deteriorated depth maps. Multiresolution fusion-based method is proposed for image focus volume enhancement. First, an initial focus volume is obtained by applying a conventional focus measure. Then, a pyramid of focus volumes is computed using Gaussian filters and subsampling. Focus measures from various focus volumes at different levels are merged into a single resultant focus volume. Finally, the depth map is obtained from the resultant focus volume by maximizing the focus measure in the optical-axis direction. According to the best of my knowledge, the cross-scale aggregation has never been used in enhancing the image focus volume in SFF. The proposed method is evaluated through the experiments using image sequences of real microscopic and simulated objects. Results comparisons based on root mean square error (RMSE) and correlation demonstrate the effectiveness of the proposed method in improving the focus volume and depth map. The proposed fusion method of volumes is a simple but effective. The idea of cross-scale aggregation of focus measures is effective in providing precise focus measures that consequently, provide accurate depth map. In future work, it will further be explored and a more sophisticated and optimization-based fusion algorithm will be applied.