高效多注意力特征融合的图像超分辨率重建算法

Efficient Multi-attention Feature Fusion for Image Super-resolution Reconstruction Algorithms

为改善图像质量,提升观测效果,针对现有超分辨率重建算法由于网络层数过深导致的信息丢失、参数量大的问题,提出一种高效多注意力特征融合的图像超分辨率重建算法(EMAFFN).该算法通过渐进式特征融合块(PFFB)逐步提取图像的特征信息,减少特征信息在深层次网络传递过程中的丢失,同时结合PFFB内部的高效多注意力块(EMAB)在通道和空间两个分支作用,自适应的对提取到的特征进行加权,使网络更多的关注高频信息,最后使用多尺度感受野块(RFB_x)对提取到的特征进行增强、并多尺度融合特征来提升重建模块的性能.实验结果表明,EMAFFN在公共数据集Set5上的平均PSNR值最高达到37.93dB,SSIM达到0.9609,重建后的图像恢复了更多的高频信息,纹理细节丰富,更接近于原始图像.

To improve the image quality and enhance the observation effect,an image super-resolution reconstruction algorithm with efficient multi-attention feature fusion(EMAFFN)is proposed to address the problems of information loss and large number of parameters caused by the existing super-resolution reconstruction algorithm due to the deep network layers.The algorithm gradually extracts the feature information of the image through the progressive feature fusion block(PFFB)to reduce the loss of feature information in the process of deep network transmission,and at the same time combines the efficient multi-attention block inside the PFFB to act in both channel and space branches,adaptively weighting the extracted features to make the network focus more on high frequency information,finally,multi-scale receptive field blocks(RFB_x)are used to enhance the extracted features,and multi-scale fusion features are used to improve the performance of the reconstruction module.The experimental results show that EMAFFN achieves an average PSNR of 37.93dB and an SSIM of 0.9609 on the public dataset Set5.The reconstructed images recover more high-frequency information and are rich in texture details,which are closer to the original images.