This study proposes a method for correcting distortion in full-spherical (omnidirectional) images, aiming to achieve high-precision depth estimation. In image processing applications, distance estimation is frequently required. Among various measurement methods, depth estimation using machine learning is employed. Depth estimation involves training models on datasets containing pre-created images and distance information to estimate relative distances within input images. However, depth estimation models trained using machine learning typically rely on images captured by conventional cameras. Consequently, distorted images, such as those from omnidirectional cameras, cause inaccurate results. The proposed method aims to achieve high-precision depth estimation for spherical panoramas by first correcting the distortion in the image, then performing depth estimation, and finally applying an inverse transformation. Experimental results from implementing the proposed method confirmed an improvement in depth estimation accuracy compared to before distortion correction.