Automatic alignment and comparison of petri dish images containing cell colonies : submitted to the Institute of Information & Mathematical sciences in partial fulfillment of the requirements for the degree of Master of Information Science in Software Engineering at Massey University

Thumbnail Image
Open Access Location
Journal Title
Journal ISSN
Volume Title
Massey University
The Author
This work proposes a novel approach comprising of a chain of algorithms, for comparing, matching and aligning pairs of cell colony images taken at different stages on a Petri dish. The objective is to provide an assistive tool for microbiologists to quantify the loss or growth of cell colonies on two Petri dishes, by mapping cell colonies between a pair of images. This problem is highly non-trivial, as the shape, size and position of the corresponding colonies vary randomly. In addition, the cell colony images for comparison were taken at different times and from slightly different perspectives (i.e. effects of shearing); therefore, amplifying the complexity of the problem. Preliminary studies show that approaches purely based on SIFT or SURF, as well as algorithms used in astronomy, do not perform well on the problem domain. We therefore introduce a new approach to addressing these problems. A novel iterative technique that combines triangulation algorithms with the RANSAC alignment algorithm and AdaBoost classifer for alignment validation is proposed. Using 60 pairs of images of Petri dishes containing real biological cell colonies, we demonstrate the efficacy of the new algorithm in comparison to existing ones found in the literature. Empirical results show that the new proposed algorithm, we call K-NT for cell colonies matching, performed 4 times more accurate than other existing triangulation-based pattern matching algorithms. In the last stage of processing, we were able to generate an AdaBoost classifer with an accuracy of 98.5% that helps validate if an image was successfully aligned or not.