Translator Disclaimer
9 June 1998 Two-dimensional defocusing correction using artificial neural nets
Author Affiliations +
Proceedings Volume 3261, Three-Dimensional and Multidimensional Microscopy: Image Acquisition and Processing V; (1998)
Event: BiOS '98 International Biomedical Optics Symposium, 1998, San Jose, CA, United States
The aim of this work is to show that properly trained ANNs can be used as an image restoration tool to correct the effect of defocusing on 2D optical microscopy images. The proposed method can be applied to correct the results of inaccurate range focusing algorithms on fully automated imaging and analysis systems and to compensate the effect of the limited of the depth of focusing on high numerical aperture system. One type of ANN was used: feedforward multilayer perceptron, with supervised back propagation training. Its performance has been tested with both synthetic images and real images from latex microspheres. The network was trained with sets of pairs of images: each pair consisted of a defocused image and its corresponding in-focus version. Different levels of defocusing were used. The criteria used to select the algorithm parameters to tune the networks and to train them will be presented. The results of the experiments performed to test their ability to 'learn' to correct the defocusing and to generalize the results will also be shown. The result show that, when trained with images with some levels of defocusing, the network was able to learn and accurately correct these defocusing levels, but it can also generalize the results and correct other levels of defocusing.
© (1998) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Carlos Ortiz de Solorzano, Vicente Gonzalez, Andres Santos, and Francisco del Pozo "Two-dimensional defocusing correction using artificial neural nets", Proc. SPIE 3261, Three-Dimensional and Multidimensional Microscopy: Image Acquisition and Processing V, (9 June 1998);

Back to Top