This paper presents an improved region growing method for the segmentation of images comprising three phases. It is shown that image segmentation errors usually occur at the interfaces between the two phases with the highest and lowest grayscale intensity levels among the three phases (i.e. partial volume effect rings appear) when global thresholding (GT) or a traditional region growing (TRG) method are used for image segmentation. The proposed method can identify the interfaces, and thereby effectively avoid the image segmentation errors. This method was validated through its application to the image segmentation of two partially saturated soils and comparison of the image-based calculation of water-content values to experimentally measured values. The soil-water (S-W) interface areas of the soils were quantified based on the image-segmentation results acquired from different methods. The results indicate that GT and TRG significantly overestimate the S-W interface area. (C) 2020 Elsevier B.V. All rights reserved.