In this project, we aim to tackle the problem of image matting, which is essentially seperating the foreground from the background of an image automatically. We use a closed-form matting solution implemented at https://github.com/MarcoForte/closed-form-matting based on the paper https://people.csail.mit.edu/alevin/papers/Matting-Levin-Lischinski-Weiss-CVPR06.pdf. We use this closed-form matting algorithm along with a trained UNET on a large human potrait dataset (https://www.kaggle.com/datasets/laurentmih/aisegmentcom-matting-human-datasets?select=matting) and compare error metrics such as MSE, PSNR, and SAD to evaluate our two methods and compare the results.
moderatelygreek / themattingmen Goto Github PK
View Code? Open in Web Editor NEWAn exploration of both neural and classical matting techniques.