论文部分内容阅读
提出了一种基于时空特征和神经网络的视频超分辨率重建算法,实现了视频视觉分辨率质量和细节清晰度的提升.该算法综合考虑了外部图像块之间的关联映射关系和内部图像块间的相似性,利用深度卷积神经网络学习得到的拟合系数快速地重建视频细节.采用时空非局部特征相似性优化重建结果,将相邻视频帧间的非局部互补冗余信息融入学习视频帧结果中,解决了误匹配等问题,进一步提升了超分辨率性能.实验结果表明,所提方法在客观评价指标和主观视觉效果上均取得了较好的重建效果.
This paper proposes a video super-resolution reconstruction algorithm based on spatio-temporal features and neural network, which realizes the improvement of video visual resolution quality and detail resolution.The algorithm considers the relationship between the external image blocks and the internal image blocks And reconstruct the video details quickly by using the fitting coefficients obtained by deep convolutional neural network learning.Using the similarity of spatio-temporal non-local features to optimize the reconstruction results, the non-local complementary redundant information between adjacent video frames is integrated into the learning video In the frame result, the problems of mismatching and other problems are solved and the super resolution performance is further improved.The experimental results show that the proposed method has achieved good reconstruction results both in objective evaluation index and subjective visual effect.