Automatically Count and Classify Multiple Colonies on Mobile Device Captured Images
Date Issued
2016
Date
2016
Author(s)
Chen, Yun-Tao
Abstract
The purpose of this study is to develop a mobile app that can automatically count and classify multiple bacteria colonies to solve the low efficiency and time consuming manual counting problems. Although there are many automatic colony counting instruments, they suffered from many limitations. Currently, mobile devices are widely spread. Therefore, the goal is to develop an mobile app which can automatically count and classify colonies. The user can take the photos and use the app to identify colonies for summing up the total number without setting any parameter. This study includes two parts, the first part is to distinguish foreground (colonies) and background (agar and others). We propose methods for isolate the noise from photographs taken by handheld device. A small region (region of interest) detection method and the color characteristics are used in proposed approach to distinguish foreground and background. The second part is to automatically classify different colony types and count total number of colonies in each types. To classify the colonies in the foreground image extracted in first part, we use the values of CIELAB color space of the image as the features, and use K-means clustering algorithm to classify these images. Furthermore, we establish the colony characteristic model for each type of colonies to rescan the uncertain areas, and sums up to the total number for every colony. In this study, we established two groups of dataset for efficacy comparison and a systematical comparison tool. The first dataset group includes 33 pictures that taken by handheld devices with many noises. The second dataset group includes 49 pictures obtained from OpenCFU open dataset with no noise. Both of the two datasets include single colony species and multiple colony species. The proposed approach achieved the followings: (1) the average F-measure of colony detection can achieve 71.6% in the first dataset, 91.7% in the second dataset and results to the overall accuracy 83.6%, (2) the average F-measure of multiple colony classification can achieve 91.5% and the average rand index (RI) values reaches 92.6%. Compare to the existing studies or tools (Colony Counter v1.0, OpenCFU, CFU Scope), the proposed approach performs the best.
Subjects
colony counting
colony classifying
image segmentation
feature extraction
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-105-R03525056-1.pdf
Size
23.54 KB
Format
Adobe PDF
Checksum
(MD5):197c80661c9e931d8303bf0f8a768d16