iDEA: Drexel E-repository and Archives >
Drexel Theses and Dissertations >
Drexel Theses and Dissertations >
Improving an automated mosaicing algorithm for wide-field microscopy
Please use this identifier to cite or link to this item:
|Title: ||Improving an automated mosaicing algorithm for wide-field microscopy|
|Authors: ||Jhaveri, Sankhesh Jayesh|
|Keywords: ||Biomedical engineering|
Microscopy -- Technique
Tissues -- Imaging
|Issue Date: ||15-Apr-2010 |
|Abstract: ||Knowledge of the complex fiber structures of soft tissues can lead to greater understanding of basic structure-function relationships, and potentially, to improvements in tissue engineered constructs and micro-repair techniques. Unfortunately, imaging these structures in fresh, whole-tissue samples is difficult, mainly because current microscopes are designed for small-scale, narrow field imaging of thin, slide-mounted specimens. The currently available wide-field microscopy systems depend on high-precision motorized (using expensive servo motors) stage positioning to make a montage of image tiles. However, they are time consuming requiring a large number of image tiles. To achieve high speed, high resolution, wide-field imaging at low cost, a novel microscopy system was developed that is capable of imaging thicker, fresh tissue samples as well as prepared slides using both, normal and polarized light. The system uses a low-precision, two stepper motor positioning system, maintaining sub-pixel accuracy via a novel image correlation and registration algorithm. A third stepper motor controlling the fine focus knob enables Z axis control for automatic focus. The software consists of two basic components: 1. A graphical user interface (GUI) programmed in Visual Basic .NET for camera and stage motor control, and 2. A “tiling/stitching” algorithm programmed in MATLAB®.
The tiling program incorporates distortion and luminosity correction algorithms. Autofocus is achieved using a novel edge-based focusing algorithm. The elementary algorithm for image stitching was limited in its use owing to its large dynamic memory requirement as well as significantly high computational time. A more challenging limitation of the algorithm was its inability to detect and account for white spaces which are feature-less ‘blank’ regions of the image. This would lead to erroneous image stitching which would render the final image ineffectual for data analysis. The objective of this work was to optimize the system algorithms to achieve dynamic and accurate image stitching using improved feature and ‘blank area’ detection. The point selection algorithm to correlate pixel windows was optimized to detect most favorable features for tracking. A novel algorithm was developed to ‘flag’ or mark the overlapping regions with white spaces (no information) while stitching and then, estimating the position of the flagged tiles in the final image. To demonstrate superior usability of the microscopy system and the stitching algorithm, we imaged distinct sets of histology slides using normal and polarized light. The optimization of system algorithms made reconstruction of montage images faster, accurate and much more reliable. The optimized algorithms also make judicious use of system resources.|
|Appears in Collections:||Drexel Theses and Dissertations|
Items in iDEA are protected by copyright, with all rights reserved, unless otherwise indicated.