Improved GrabCut Algorithm for Classify Mycobacterium Tuberculosis

Abstract

Feature extraction is a stage in the image processing process. There are many feature extraction methods used for image processing, one of which is foreground extraction. Foreground extraction, which is image segmentation, is needed to separate the main object in the image that will be processed in image processing. It is needed to select the main object from the background so that the image processing process can focus on the main object. Several algorithms can be applied to perform foreground extraction, one of the most popular is the GrabCut algorithm. In this article, we propose a method for classifying Mycobacterium Tuberculosis as the main object by adding the yolov8 architecture to perform foreground extraction and classification of microscopic images of tuberculosis. In this article, foreground extraction is carried out using the GrabCut algorithm and adding the yolov8 architecture using bounding boxes. The Mycobacterium Tuberculosis data used was 1265 images. The proposed method can classify and calculate the number of Mycobacterium Tuberculosis. The best results on the validation dataset are a method using Yolov8, with a data partition of 90:10 which produces an accuracy value of 82%, a precision value of 99%, a Recall value of 82.9%, an mAP value of 80% and a MAPE value of 6.1% which has a very accurate interpretation of forecasting results