Coarse Segmentation of Textured Images Using Variance Analysis. | ||||
| MEJ- Mansoura Engineering Journal | ||||
| Article 5, Volume 22, Issue 3, September 1997, Page 1-12 PDF (157.18 K) | ||||
| Document Type: Research Studies | ||||
| DOI: 10.21608/bfemu.2021.150960 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
| Fatma Samir* 1; Ayman A.M. Ibrahim2; Samir I. Shaheen3 | ||||
| 1National Telecommunication Institute., 5 Mokhaiam El-Dam Street., Naser City., Cairo., Egypt. | ||||
| 2National Telecommunication Institute.,5 Mokhaiam El-Dam., Cairo., Egypt. | ||||
| 3Computer Engineering Department., Faculty of Engineering., Cairo University., Giza., Egypt. | ||||
| Abstract | ||||
| This paper presents a novel approach for the segmentation of a textured scene. The algorithm is image-based; no specific model is assumed for the image. Also, no a priori knowledge about the different texture regions, neither their number, nor their behavior is assumed. The algorithm partitions the image into small disjoint square windows, and the Variance for each window data is calculated. Then the K-means clustering algorithm is applied upon these windows. Together with the K-means algorithm, a new distance measure has been defined. This new distance measure was deduced from a statistical test known as Bartlett's test based on the variance of the window's data. The same statistical test has also been applied but in a different fashion to determine the number of different textures in the image. The new image-based distance measure has been tested and compared to a model-based Euclidean distance measure, with each window modeled by a non-causal Gaussian Markov Random Field (GMRF). The results of the comparison have shown that the new distance measure is much simpler and faster, while yielding to a still robust and effective segmentation. | ||||
|
Statistics Article View: 72 PDF Download: 144 |
||||
