Abstract:
Although selfie images have become popular among smartphone users, the supportive hand that holds the camera ruins the beauty of the picture. The captured images will look more realistic if the supporting hand is removed from the original image. This paper proposes a machine learning and a computer vision based approach to remove the supportive hand and reconstruct the removed hand that matches with the person who took the picture. A fully convolution neural network (FCN) and a partial convolution neural network (PConv Net) have been used to accomplish this task. Results indicate that the FCN gives 94.58% validation accuracy with the PConv Net is utilized to train the model for background matching and hand creation. The FCN and PCovNet models minimize validation lose up to 1.73 and 1.95, respectively.
Citation:
W. M. H. I. Weerakoon and R. G. N. Meegama, "Convolutional Neural Network to Reproduce Selfie Images after Removing Supportive Hand," 2021 6th International Conference on Information Technology Research (ICITR), 2021, pp. 1-6, doi: 10.1109/ICITR54349.2021.9657454.