Identificación de parásitos de reptiles usando técnicas de procesamiento de imágenes, visión por computador y aprendizaje automático : segmentación de imágenes de parásitos de reptiles mediante técnicas de procesamiento digital de imágenes.

Abstract

The present integration work implements the segmentation of reptile parasite images using MATLAB software through three segmentation techniques. Initially, the extraction and analysis of the mask for each of the images were performed, which were obtained from a large database of the research project "Identification of parasites using different coprological methods in reptile samples at the Quito Vivarium." The project includes 3616 images, and its objective was to distinguish gastroenteric parasites in reptiles.

The Image Labeler tool, which was used, was added to the MATLAB catalog in version 2022b. This tool facilitates the acquisition of labels that mark or highlight the region of interest (ROI). The corresponding masks for each image were divided into two groups: 70% for training, which is approximately 2532 images, and 30% for validation. These images are randomly selected each time the program and/or code is run, meaning they change.

The implemented techniques were Otsuthresh, Activecontour, and Imsegfmm, which work with 2D images and serve the purpose of image training. Using the training images and the application of each method, the segmentation metric called "DICE" was analyzed. DICE is the Sørensen-Dice similarity coefficient for image segmentation. The best average was obtained through pre-training to observe which of the three techniques achieves the best efficiency. In other words, the best accuracy for image segmentation was evaluated.