Spatial Discriminant Function with Minimum Error Rate for Image Segmentation
This paper describes how a normal discriminant function with minimum error rate can be applied to segment an image in a particular manner. Since the maximum likelihood method assigns pixels based on the underlying distributions in image, it is inevitable to make decision errors when there are overlapping areas between the underlying distributions. However, this overlapping area can be minimized by a conversion of distributions which is proposed in this paper. This method is derived by exploiting characteristics of a linear combination of random variables and its relation to the corresponding random vector. The suitable performance of the process is mathematically proved and the experimental results that support the effectiveness of the proposed method are provided.
KeywordsImage Segmentation Random Vector Discriminant Function Neighboring Pixel Decision Error
Unable to display preview. Download preview PDF.
- 5.Sakai, M., Yoneda, M., Hase, H.: A New Robust Quadratic Discriminant Function. In: Proc. Int’l. Conf. Pattern Recognition (ICPR1998), pp. 99–102 (1998)Google Scholar