Shaft orbit recognition is an important approach for the vibration state judgment of steam turbines. Extracting the features of shaft orbit images is not an easy task, and the traditional feature extraction methods are not perfect in comprehensiveness, accuracy and stability. In order to overcome these problems, a feature extraction method based on imitating human eyes is proposed for the steam turbine. This method imitates human eyes to extract the most important information of the image structure, boundary and region, and realizes the shape characterization comprehensively and accurately through full integration of the information. Three intelligent classification methods are used to test the effectiveness of the proposed method, and the experimental results prove that this feature extraction method for steam turbine is simple, efficient and accurate.