In medical imaging, denoising is very important for the analysis of images and the diagnosis and treatment of diseases. Currently, the image denoising methods based on anisotropic diffusion are efficient. However, the methods have been limited as regards the processing time. In recent computing systems, the FPGA-based acceleration has been highly competitive for GPU-based one due to its high computation capabilities and lower energy consumption. In this paper, we present a high-level synthesis implementation on a SOC-FPGA of an anisotropic diffusion algorithm dedicated to medical applications. We choose an oriented speckle reducing anisotropic diffusion denoising filter, which provides robust performance but requires a significant computation on the embedded CPU since it is iterative. Moreover, we optimize the performance by modifying the original algorithm, automizing it by controlling the diffusion process at each iteration, and accelerating the processing operations by providing a hardware/software description. The evaluation is performed using different medical images. The efficiency and relevance of the proposed filter is demonstrated through segmentation. The design is validated on FPGA XC7Z020CLG484-1 with a frequency of 255 MHz and a PSNR of about 30 dB.

在医学成像中，去噪对于图像的分析和疾病的诊断和治疗非常重要。目前，基于各向异性扩散的图像去噪方法是有效的。然而，这些方法在处理时间方面受到限制。在最近的计算系统中，基于 FPGA 的加速由于其高计算能力和较低的能耗，在基于 GPU 的加速方面具有很强的竞争力。在本文中，我们在 SOC-FPGA 上展示了专用于医疗应用的各向异性扩散算法的高级综合实现。我们选择了定向散斑减少各向异性扩散降噪滤波器，它提供了强大的性能，但由于它是迭代的，因此需要在嵌入式 CPU 上进行大量计算。而且，我们通过修改原始算法来优化性能，通过控制每次迭代的扩散过程使其自动化，并通过提供硬件/软件描述来加速处理操作。使用不同的医学图像进行评估。通过分割证明了所提出的过滤器的效率和相关性。该设计在 FPGA XC7Z020CLG484-1 上进行了验证，频率为 255 MHz，PSNR 约为 30 dB。