To shorten the time for designing and developing new materials, fast evaluation of microstructure-property relation is required. In this research, we applied machine learning to perform crystal phase segmentation and extracted microstructure features of Cr-based duplex alloys from the scanning electron microscope (SEM) images. The results show that the accuracy of crystal phase segmentation was improved when the backscattered electron (BSE) images with controlled channeling contrast was used. This indicates that measurement conditions of the BSE images are important for obtaining high segmentation accuracy. The segmented images were used for the extraction of microstructure features. With the extracted features, the reliable prediction of mechanical properties of Cr-based duplex alloys was verified with a prediction error less than 10%. We demonstrated a method of using microstructure features extracted from SEM images for fast evaluation of microstructure-property relation.