Liver segmentation from abdominal Com- puted Tomography (CT) images aims to separate the liver from other organs. It plays an important role in determining liver function and researching liver diseases. However, manual segmentation is time-consuming and labor-intensive for radiologists. The segmentation task is mitigated with emerging deep learning algorithms, and the liver region is effectively automatically extracted. In addition, images need to be appropriately processed before being used in learning algorithms. This study examines the effects of denoising techniques like 3D Median and 3D Bilateral filters on medical image segmentation tasks performed by U-Net and Dense-UNet algorithms. First, we apply one of two denoising techniques - 3D Median filter and 3D Bilateral filter - to the original CT im- ages. Meanwhile, the manual liver masks with ten-level are converted to binary masks. Next, the dimension of each denoised CT slice and binary mask are normalized from 512×512 to 256×256. Finally, we conduct the model training based on U-Net and Dense-UNet architectures to perform the segmentation on denoised CT slices. The experimental results show that using a denoising technique such as Bilateral can benefit accuracy and inference time in liver segmentation tasks on CT images.