软硬件协同的图像放大系统设计

Software and Hardware Collaborative Image Enlargement System Design

图像放大应用场景广泛,使用插值的图像放大计算速度块,但基于插值的放大处理放大效果一般,基于卷积神经网络模型的图像放大效果优秀,但是处理速度不够,本文采用一种简易的双层模型实现图像放大,并通过FPGA设计实现双层模型硬件加速器,在PYNQ-Z1板卡上通过软件方式调用双层模型加速器进行图像放大,从而实现了软硬件协同的图像放大系统,本文设计的图像放大系统比三次卷积插值处理速度快22%,放大效果上PSNR高0.76,有着明显的视觉提升效果。

Image enlargement has a wide range of applications.Interpolation-based algorithms for image enlargement are fast,but generally provide average scaling results.On the other hand,image enlargement based on convolutional neural network(CNN)models is excellent,but the processing speed is not fast enough.In this paper,we propose a simple tow-layer model for image enlargement and design an FPGA-based hardware accelerator for the two-layer model.The accelerator is called using software on the PYNQ-Z1 board to achieve image enlargement,resulting in a software-hardware co-design image enlargement system.Our designed image enlargement system is 22%faster than third-order convolutional interpolation in terms of processing speed,and provides a PSNR improvement of 0.76,resulting in a significant visual enhancement.