Open AccessAutomated Segmentation of Leukocytes using Marker-based Watershed Algorithm from Blood Smear Images
Author(s) -
Vipasha Abrol,
Sabrina Dhalla,
Jasleen Saini,
Ajay Mittal,
Sukhwinder Singh,
Savita Gupta
Publication year - 2021
Publication title -
aijr proceedings
Language(s) - English
Resource type - Conference proceedings
ISSN - 2582-3922
DOI - 10.21467/proceedings.114.9
Subject(s) - computer science , image segmentation , artificial intelligence , segmentation , thresholding , image processing , otsu's method , computer vision , scale space segmentation , rgb color model , process (computing) , segmentation based object categorization , hsl and hsv , pattern recognition (psychology) , image (mathematics) , medicine , virus , virology , operating system
The aim of this paper is to perform segmentation of white blood cells (WBCs) using blood smear images with the help of image processing techniques. Traditionally, the process of morphological analysis of cells is performed by a medical expert. This process is quite tedious and time consuming. The equipments used to perform the experiments are very costly and might not be available in all hospitals. Further, the whole process is quite lengthy and prone to error easily because of the lack of standard set of procedure. Hence there is a need for innovative and efficient techniques. An automated image segmentation system can make the blood test process much easier and faster. Segmentation ofa nucleus image is one of the most critical tasks in a leukemia diagnosis. In this work, we have investigated and implemented image processing algorithms to segment cells. The proposed model detects WBCs and converts cell images from RGB to HSV color space using Otsu thresholding. The resultant image is then processed with the morphological filter because the segmented image contains noise which affects the system performance. Lastly, the Marker-based watershed algorithm is implemented in which specific marker positions are defined. The proposed model is tested on publically available ALL-IDB2 dataset. The system’s performance was overall examined and resulted in 98.99% overall precision for WBC segmentation.