基于改进TSRN的图像超分辨率重建算法

Image Super-resolution Reconstruction Algorithm Based on Improved TSRN

针对常用图像超分辨率重建算法生成的超分辨率(SR)图像质量较差的问题,提出一种基于改进文本超分辨网络(TSRN)的图像超分辨率重建算法。在TSRN基础上加入信息蒸馏块(IDB),IDB中的增强单元使长短路径特征在通道中合成以加强浅层特征,在TextZoom数据集上训练和测试,实验结果表明,本算法可把低分辨率(LR)图像转化为更清晰的SR图像,相比其他超分辨率重建算法,其生成的SR图像更清晰,重建效果更佳。使用卷积循环神经网络(CRNN)模型对LR图像和SR图像进行文字识别,实验结果表明,本算法生成SR图像的平均文字识别准确率相较LR图像提升了16.8%,提升效果显著,且较原TSRN算法也有所提高,证明了本算法的有效性。

Aiming at the problem of poor quality of super-resolution(SR)images generated by image super-resolution reconstruction algorithms,an image super-resolution reconstruction algorithm based on improved text super-resolution network(TSRN)is proposed.An information distillation block(IDB)is added to TSRN,and the enhancement unit in the IDB makes the long and short path features synthesized in the channel to enhance the shallow features.Training and testing on the TextZoom dataset,the experimental results show that the proposed algorithm converts the low-resolution(LR)image into a clearer SR image.Compared with the previous super-resolution reconstruction algorithm,the SR image generated by the algorithm in this paper is clearer.Reconstruction is better in quality.The convolutional recurrent neural network(CRNN)model is used to perform text recognition on LR images and SR images.The experimental results show that the average text recognition accuracy of the SR image generated by the proposed algorithm is 16.8%higher than that of the LR image,and the improvement effect is significant.Moreover,it is also improved compared with the original TSRN algorithm,which proves the effectiveness of the proposed algorithm.