An Efficient Microbes Detection System using Microscopic Images via Morphological and Correlation Based Features

Anaahat Dhindsa^1,3, Sanjay Bhatia², Sunil Agrawal³ and B.S. Sohi⁴

¹Department of Electronics and Communication Engineering, Chandigarh University, Gharuan, Punjab, India,140413, 0000-0003-3877-6851

²Department of Zoology, University of Jammu, India, 180016

³UIET, Panjab University, Chandigarh, India, 160014

⁴Chandigarh University, Gharuan, Punjab, India, 140413

Corresponding Author E-mail : anaahat.dhindsa85@gmail.com

Abstract: This research work is motivated by a need to focus on computational biodiversity studies, to contribute towards research in maintaining the ecological balance of the earth. The field of computational biodiversity can leverage current advents in image processing and machine learning algorithms. This paper gives information on how to develop a pipeline of algorithms that can support biodiversity studies. The process of manual identification of algae in water bodies is tedious and laborious, and highly dependent on experts. The work demonstrated here provides methods to resolve this problem by automating the process. A hybrid algorithm that uses pixel clustering and Kirsch filter for extracting the bodies of microbes from images has been developed with high accuracy. For the automatic identification process, extensive study is done on comparing Classification and Regression Trees (CART), K-nearest- neighborhood, Gaussian Naive Bayes, Linear Regression, Linear discriminant analysis and Support vector classifier (SVC) methods. This study shows that CART algorithm is the most stable and consistent performer which is evident from the values of recall, precision, and accuracy. The SVC algorithm was second in performance.

Keywords: Algae; Classifiers; Computational Biodiversity; Image Segmentation

[ HTML Full Text]

Back to TOC