Refined Background Matting Technology Based on Implicit Neural Representation

Background matting technology is widely used today, and numerous studies have been conducted in this field. Recent research has revealed that although current background matting frameworks have increased processing speed, the quality of their output results still need improvement by solve the inaccurate edge detection and hard to handling complex backgrounds. In order to achieve more accurate results, we propose a refinement module based on implicit neural representation (INR, model complex relationships between variables in data automatically) for existing background matting frameworks. This technology combines an INR approach with a fully convolutional neural network (FCNN) to automatically label each pixel with its foreground and background attributes. We tested our technique using some matting datasets, and the results demonstrate superior accuracy to traditional methods. Therefore, this paper introduces a new background matting technology based on implicit neural representation that strives to deliver improved output accuracy, and the proposed technology provides a new way to solve the challenging task of background matting.