Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightExperimental results revealed that the proposed idea regularizes the energy terms more effectively and improves the segmentation results significantly.
📌 3 Maddede Bu Makale
Özet
Graph cut minimization formulates the image segmentation as a linear combination of problem constraints. The salient constraints of the computer vision problems are data and smoothness which are combined through a regularization parameter. The main task of the regularization parameter is to determine the weight of the smoothness constraint on the graph energy. However, the difference in functional forms of the constraints forces the regularization weight to balance the inharmonious relationship between the constraints. This paper proposes a new idea: bringing the data and smoothness terms on the common base decreases the difference between the constraint functions. Therefore the regularization weight regularizes the relationship between the constraints more effectively. Bringing the constraints on the common base is carried through the statistical significance measurement. We measure the statistical significance of each term by evaluating the terms according to the other graph terms. Evaluating each term on its own distribution and expressing the cost by the same measurement unit decrease the scale and distribution differences between the constraints and bring the constraint terms on similar base. Therefore, the tradeoff between the terms would be properly regularized. Naturally, the minimization algorithm produces better segmentation results. We demonstrated the effectiveness of the proposed approach on medical images. Experimental results revealed that the proposed idea regularizes the energy terms more effectively and improves the segmentation results significantly.
Abstract
Graph cut minimization formulates the image segmentation as a linear combination of problem constraints. The salient constraints of the computer vision problems are data and smoothness which are combined through a regularization parameter. The main task of the regularization parameter is to determine the weight of the smoothness constraint on the graph energy. However, the difference in functional forms of the constraints forces the regularization weight to balance the inharmonious relationship between the constraints. This paper proposes a new idea: bringing the data and smoothness terms on the common base decreases the difference between the constraint functions. Therefore the regularization weight regularizes the relationship between the constraints more effectively. Bringing the constraints on the common base is carried through the statistical significance measurement. We measure the statistical significance of each term by evaluating the terms according to the other graph terms. Evaluating each term on its own distribution and expressing the cost by the same measurement unit decrease the scale and distribution differences between the constraints and bring the constraint terms on similar base. Therefore, the tradeoff between the terms would be properly regularized. Naturally, the minimization algorithm produces better segmentation results. We demonstrated the effectiveness of the proposed approach on medical images. Experimental results revealed that the proposed idea regularizes the energy terms more effectively and improves the segmentation results significantly.