An Apparel Image Segmentation Method Utilizing a Prior Probability Mask Graph

Image segmentation is a potent technique for isolating clothing areas in images, particularly when dealing with complex backgrounds in clothing images. Currently, research in image segmentation predominantly revolves around deep learning, with convolutional neural networks being a prevalent choice. However, in cases involving intricate clothing deformations and edges, the segmentation outcomes are less than ideal. To enhance garment image segmentation performance, this paper presents an approach rooted in prior probability mask graphs. It introduces a multi-scale clothing keypoint fusion model, integrating supplementary clothing keypoint information into the target image. This process optimizes the data for the prior probability mask graph and enables automatic clothing image segmentation in conjunction with SAM. Following testing on the DeepFashion2 dataset, our model attained a 94% Pixel Precision and a 72% Intersection over Union.