Abstract: In recent years,computer vision has become the focus of research in various disciplines and has been gradually applied to numerous scientific research scenarios.Medical workers often use blood cell image analysis systems to automatically count and classify white blood cell images when performing blood routine tests in the clinic.Among them,the white blood cell image quality affects the counting classification effect of the blood cell analysis system.This paper focuses on the problem of blurred details of white blood cell images under the microscope and attempts to introduce a super-resolution method to solve the problem.This method introduces a Residual-in-Residual Dense Block(RRDB) based on the Super-Resolution Generative Adversarial Network(SRGAN) to improve the network structure and remove the batch normalization layer in the standard residual block.The network performance is improved and the loss function of the discriminator is improved.Experimental results show that,compared with 3 interpolation methods and 4 learning-based super-resolution methods,the proposed method(SRGAN+) improves the reso-lution while obtaining images with richer textures and more realistic visuals.Compared with the SRGAN method,the proposed algorithm has a 1.008 dB improvement in peak signal-to-noise ratio(PSNR) and 1.07% improvement in Structural SIMilarity(SSIM).

Key words: Generative adversarial network, Leukocyte image, Residual-in-Residual dense block, Super-resolution