Multi-objective evolutionary algorithms based operation sequence design for image segmentation

Abstract

Image segmentation transforms an image into a more understandable representation by grouping pixels with common characteristics, making it easier to identify regions of interest. There is no optimal segmentation method for all cases, which makes it challenging to select the appropriate technique for each image. We propose using a Multi-Objective Evolutionary Algorithm (MOEA) to generate sequences of operations adapted to specific applications. The evolutionary process was guided by objective functions based on Receiver Operating Characteristics (ROC) Curve analysis, using maximization of sensitivity and specificity. The experiments were performed with three types of images: cells (type A, B, C, and D), melanoma images (benign and malignant), and retinal ophthalmoscopic images. The results show that the algorithm achieves a sensitivity (TPR) and specificity (TNR) of up to 1.0 in the segmentation of images of type A and D cells, a sensitivity of 1.0 and specificity of 0.9765 in the segmentation of images of benign melanoma, a sensitivity of 0.9857 and specificity of 0.9825 in the segmentation of malignant melanoma images, and sensitivity of 0.8931 and specificity of 0.9104 in the extraction of retinal veins in ophthalmoscopic images.