deinterlacing; back propagation artificial neural network; image format conversion
Issue Date
2013-07
Publisher
SPIE - THE INT SOC FOR OPTICAL ENGINEERING
Citation
Optical engineering, 2013, 52(7), P.073107, 8P.
Abstract
A back propagation artificial neural network (BP-ANN) has good self-learning, self-adaptation and generalization abilities, which we intend to use to interpolate an image. The interpolated pixels are classified into two regions, each region corresponding to one BP-ANN. In order to optimize the structure of the BP-ANN and the process of deinterlacing, three experiments were performed to test the architecture and parameters of region-based BP-ANN. The experimental results show that the proposed algorithm with an 8 - 16 - 1 structure provides the best balance between time consumption and visual quality. Compared to the other six advanced deinterlacing algorithms, our region-based BP-ANN method provides about an average of 0.14 to 0.64 dB higher peak signal-to-noise-ratio while maintaining high efficiency. (c) 2013 Society of Photo-Optical Instrumentation Engineers (SPIE)