แอ็กทิฟคอนทัวร์แบบใช้ข้อสนเทศบริเวณท้องถิ่นบนเส้นค้นหาที่ยืดได้สำหรับการแบ่งส่วนภาพ / โสภณ ผู้มีจรรยา = Active contour using local regional information on extendable search lines for image segmentation / Sopon Phumeechanya
This dissertation presents a novel active contour method for image segmentation using local regional information on extendable search line, or the LRES active contour. Our active contour consists of a set of search lines that are perpendicular to the contour front. Moreover, the intensity profiles on search lines are calculated, so it can inform the contour front to move toward the object's boundary. When this local regional information is not enough, the length of each search line is to increase gradually until a boundary of the object is found. We compare the performance of our LRES active contour to other existing active contours, both edge-based and region-based. In the experiments, we test on synthetic images, real scene images, and images from IBSR and MMU databases. The results show that our method provides more desirable segmentation outcomes, particularly on some images where other methods may fail. The advantages of LRES are a large capture range, an ability to reach into deep concave shape, no existence of the saddle and stationary point problems, an ability to segment an image corrupted by Gaussian noise, no trapping at any sink point, and an ability to trace an object with weak boundary and heterogeneous texture. Furthermore, our LRES active contour can be extended to include an ability to segment multiple objects within the image at the same time and select the desired object using its edge type.
