一种基于立体注意力机制的立体图像超分辨算法  

A Stereo Image Super-Resolution Algorithm Based on Stereo Attention Mechanism

作　　者：罗传未 张子慧 贺子婷 周孟颖 马健 LUO Chuanwei;ZHANG Zihui;HE Ziting;ZHOU Mengying;MA Jian(School of Internet,Anhui University,Hefei 2300001,China)

机构地区：[1]安徽大学互联网学院,安徽合肥2300001

出　　处：《电视技术》2023年第1期30-35,共6页Video Engineering

摘　　要：针对因图像采集系统或采集环境本身的限制导致的立体图像模糊、质量低下、感兴趣区域不显著等问题,在最新的基于立体注意力模块的立体图像超分辨算法的基础上,通过在单图超分辨(Single Image Super-Resolution,SISR)的深度网络中引入立体图像左右两个视点间的互补信息以及平滑损失(Smoothness Loss)函数,增强超分辨后立体图像的视觉效果。在该算法中,梯度更小、更加平滑的立体注意力图可以获得更好的立体图像超分辨效果。为证明引入的函数有效,对改进前后的基于立体注意力机制的立体图像超分辨算法进行对比实验和分析,结果表明,引入平滑损失后,SRCNN和SRResNet模型的峰值信噪比(Peak Signal to Noise Ratio,PSNR)值和结构相似性(Structural Similarity,SSIM)值有明显提高。To address the problems of blurred stereo images, low quality and unremarkable regions of interest due to the limitations of the image acquisition system or the acquisition environment itself. In this paper, based on the latest stereo image super-resolution algorithm based on stereo attention module, the complementary information between the left and right viewpoints of stereo image and the smoothness loss function are introduced into the depth network of Single Image Super-Resolution(SISR) to enhance the stereo image after super-resolution. visual effect after super-resolution. In this algorithm, a smaller gradient and smoother stereo attention map can obtain a better stereo image super-resolution effect. In order to prove the effectiveness of the introduced function, this paper conducts comparison experiments and analysis on the stereo image super-resolution algorithm based on stereo attention mechanism before and after the improvement, and the results show that the Peak Signal to Noise Ratio(PSNR) and Structural Similarity(SSIM) values of SRCNN and SRResNet models are significantly improved after the introduction of smoothness loss.

关 键 词：图像超分辨 立体图像 立体注意力 平滑损失函数 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]