Image Segmentation by Normalized Cut with Shape Information

• Main Authors: Guo-Wei Lin, 林國偉
• Other Authors: Chin-Chun Chang
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/89420514405612604883

碩士 === 國立臺灣海洋大學 === 資訊工程學系 === 93 === Image segmentation is a classical problem in compute vision. In the recent years, some researches regard the image segmentation problem as a graph-partitioning problem. Among various graph-partitioning algorithms for image segmentation, of particular interest in this thesis is the normalized cut because the normalized cut is capable of establishing the relationship between each pair of pixels. However, to our knowledge, all of the graph-partitioning approaches only utilize low-level information about the image. In this thesis, in order to find the contour of the target shape with shape deformations, we propose a new scheme to incorporate high-level information about the target shapes, which is collected by the generalized Hough transform (GHT), into the normalized cut. The experimental results show that our approach can segment out the target shape. In addition, in comparison with the GHT, the proposed approach has better edge continuation, could tolerate larger shape variation, and cover less erroneous contours.