An Automatic Image Segmentation Algorithm Based on Three-Way Decisions

Image segmentation plays an important role in pattern recognition and computer vision. However, several problems not currently addressed affect the performance in segmenting objects from the background. First, color images tend to be divided into small patches, called over-segmentation. Second, most image segmentation algorithms are two-way decisions models, which will cause the inaccurate results when available information is insufficient. To address these problems, a three-way decisions based image segmentation algorithm is proposed in this study. Instead of dividing images into foreground and background as other methods have done, the proposed algorithm enables non-commitment for uncertain pixels and re-grows them using the information from neighboring regions. Due to the capability of reducing uncertainty and imprecision, three-way decisions is an effective method for image segmentation. The experimental results using PASCAL VOC 2007 demonstrate the effectiveness of the proposed algorithm. In addition, adopting the proposed algorithm in applications like CT image analysis results in a significant performance boost.