An edge direction based neural network interpolator for video deinterlacing
Xianglin Wang, Yeong Taeg Kim · 2003
This paper presents an image interpolation method for video deinterlacing based on edge directions and linear neural networks. Edge directions are detected by checking vector correlations between every two neighboring lines in an interlaced video field. Based on detected edge directions, new pixels are interpolated through linear neural network interpolators. For each different edge direction, a neural network is trained and used for interpolating pixels that have the same edge direction at their locations. Compared with conventional non edge direction based image interpolation method, the method presented in this paper gives clearly better edge quality in the interpolated image without introducing any obvious artifacts. In addition, due to the simplicity of linear neural network structure, the proposed method is well suited for real-time implementation.