Colour segmentation of Gram-Negative bacteria using graph Quadratic Form and Random Walker

Abstract

Abstract Gram-negative bacteria are one of the bacteria that are often pathogenic to the human body. This bacterium causes resistance due to nosocomial with other Gram-negative bacteria. In the medical stage, the bacteria that cause nosocomial traits removed first before antibiotic therapy carried out on the main bacteria. To identify these bacteria, the clinical laboratory needs to make manual observations under a microscope. The approach taken in this research is using the image processing technique. There are four stages: pre-processing, segmentation, feature extraction, and identification. Segmentation is a stage to emphasize the object sought in an image. In this research, the approach used to capture objects is one of them using the Graph Quadratic Form algorithm. This algorithm chose because it can determine the shortest distance of the object from the nearest node so that the process of convergence of the object search becomes faster. The result is that this algorithm is better than the morphology-based algorithm and the contour-based algorithm, while the number of samples taken from 50 patients affected by Gram-negative bacteria. The image under research has a size of 512x512 pixels, a resolution of 72 dpi with a bit depth of 24. The segmentation process is carried out on Gram-negative bacterial images using two classes producing an average accuracy of 89% to Ground truth.