Study of foreground extraction image segmentation techniques

Abstract

The field of computer vision has advanced significantly in the current era of technology. Digital image processing has undergone significant advancements. Image Generation, Image Enhancement, and Image Restoration are the three categories into which digital image processing techniques are commonly divided. Segmentation Procedure is one of several stages in image processing that divides an image into its individual components or objects [1]. Image segmentation is a technique that divides a digital image into a number of smaller groups or regions known as image segments, thereby lowering the complexity of the image and facilitating easier image processing and analysis [3]. There are numerous image segmentation techniques that are based on the two fundamental concepts of similarity/region and discontinuity/boundary. Techniques for image segmentation are frequently utilized in biomedicine, document processing, object recognition, automated industrial production, computed tomography (CT) images, and many other fields. Here, the goal is to compare and learn about fundamental image pre-processing and accessible segmentation approaches. ML-based algorithms are among the most effective foreground extraction methods currently available for masking images. However, it is discovered that ML approaches are unreliable. According to edges and appearance models, graph-based methods like Graphcut/Grabcut can successfully recover the foreground [6]. Both the foreground and the background are modelled using the Gaussian Mixture Model (GMM). However, with these graph-based techniques, manual creativity is necessary. The main issues with current techniques for foreground extraction in image segmentation are accuracy and performance.