Augmented Pre-Segmentation Method for Medical Image Dataset Based on Machine Vision

At present, the preparation of data is a costly and time-intensive process in in the deep learning tasks of medical images. At the same time, there is more noise in labeling, and the time cost of labeling is relatively high. We propose a method based on machine vision to reproduce multiple valid samples from original samples. In the process, the morphological feature information of the image is first identified and exacted from the medical image. Secondly, a priori features are added to divide the pictures while improving the sample availability rate. Third, the Roberts quality evaluation score is calculated to exclude low-quality samples. The example presented in the experiment shows that the sample dataset was increased up to 50-100 times the original through image processing on laparoscopic vascular images. The samples reproduced by our method can also be marked with the thick label of the original image.