Super-resolution reconstruction of video sequences based on wavelet-domain spatial and temporal processing

摘要

Super-resolution (SR) have gained increasing attention recently, especially in the fields of medical images, satellite images, and video surveillance applications. It is our focus to explore the SR techniques for videos containing diverse kinds of motions (global or local motions). To this aim, we adopt the strategy of wavelet-domain processing, focusing on the estimation of high-frequency (LH, HL, and HH bands) wavelet coefficients from both the spatial and temporal information. The detail wavelet coefficients are estimated from the reference frame by performing weighted motion compensation in the wavelet domain, which is then fused with those obtained from Lanczos interpolation of the current frame. Experiments show that our algorithm outperforms traditional interpolation methods by 1.0dB ∼ 2.2dB in reconstruction PSNR within reasonable computing time.