基于多尺度密集连接和U-Net改进的动态场景去模糊算法

针对现有去运动模糊网络在图像恢复过程中出现的纹理细节丢失、无法抑制噪声、产生振铃伪影等问题，提出一种基于多尺度密集连接和U-Net改进的动态场景去模糊算法。首先，借助U-Net网络中空洞卷积下采样有效扩大感受野，在不增加参数量的情况下避免图片产生不可逆损伤，并利用亚像素卷积在上采样过程中以小的卷积核获得清晰的图像细节，降低运算复杂度；其次，设计多尺度密集特征提取模块(multi-scale dense feature extraction, MDFE),通过密集连接的卷积层加强深层次特征提取和复用，运用空间金字塔池化(spatial pyramid pooling, SPP)分支引导多尺度特征的传递和融合，促进图像细节纹理的有效保留；最后，采用ConvLSTM双向连通结构(bidirectional convolution LSTM unit, BCLU)以非线性方式从编码路径补偿简单级联流失的上下文特征，推动深度特征跨阶段相互作用，弱化边缘伪影和噪声干扰。与现有先进方法对比，验证了本文所提算法在性能上的优势。

Dynamic scene deblurring algorithm based on multi-scale dense connection and U-Net improvement

A dynamic scene deblurring algorithm based on multi-scale dense connections and U-Net improvement is proposed to address the issues of texture detail loss,inability to suppress noise,and generation of ringing artifacts in existing motion blur removal networks during image restoration.First,the receptive field is effectively expanded by using the hole convolution downsampling in the U-Net network to avoid irreversible damage to the image without increasing the number of parameters,and the sub-pixel convolution is used to obtain clear image details with a small convolution kernel in the upsampling process,reducing the computational complexity;Secondly,a multi-scale dense feature extraction (MDFE) module is designed to enhance deep level feature extraction and reuse through densely connected convolutional layers.Spatial pyramid pooling (SPP) branches are used to guide the transfer and fusion of multi-scale features,promoting effective preservation of image details and textures;Finally,the ConvLSTM bidirectional connectivity structure is used to compensate for contextual features of simple cascading loss from the encoding path in a nonlinear manner,promoting cross stage interaction of deep features,and weakening edge artifacts and noise interference.Compared with existing advanced methods,the performance advantages of the algorithm proposed in this paper have been verified.