The CA(Channel-wise Attention)is introduced into the Basic VSR model to realize the learning of the nonlinear de-pendency of residual convolutional structure on each channel in the bidirectional information propagation process,and,there-fore,improve the video quality of surveillance video,which is essential for accurate license plate recognition.Compared with EDVR-L and conventional BasicVSR,the developed model can improve the PSNR(Peak Signal-to-Noise Ratio)by about 1.3 dB and 0.3 dB respectively as for restoring surveillance video is concerned.A license plate recognition experiment for a given traf-fic surveillance scenario is designed which demonstrates the improvement of the license plate recognition success rate gained by the new algorithm.
video super resolutionBasicVSR modelCA(Channel-wise Attention)license plate recognition
NETL
NSTL
万方数据
