Texture segmentation by clustering of Gabor feature vectors
We approach the texture segmentation problem by clustering feature vectors created from a Gabor transform data block. Given an N {times} N image, we compute 24 Gabor transforms using Gabor kernels with six orientations and four sizes. This results in a Gabor data block composed of N{sup 2} feature vectors of length 24. Each feature vector contains frequency and orientation information characteristic of its corresponding pixel in the image. In order to choose the most important feature vectors and reduce the computational complexity of the problem, we use a clustering algorithm (a variation of Kohonen's Feature Map algorithm) to group the vectors based on their distribution in feature space. We hypothesize that the pixels in a given group have similar characteristics, so that the technique can be used for texture segmentation. By clustering Gabor features, we are able to segment an image into regions of uniform texture without prior knowledge of the types of texture, or the frequency and orientation characteristics of these features. Texture segmentation algorithms that take the statistical approach and the structural approach require that we divide the image into fixed-size tiles and assume constant texture within each tile. Our approach avoid this requirement. A time complexity analysis shows that using the Kohonen algorithm for clustering is faster that the nearest neighbor algorithm and the minimum spanning tree algorithm. The Kohonen algorithm is also more stable than the k-means algorithm, as it is insensitive to the initial assignment of cluster centers. We describe experiments using both simulated and measured image, in which the resulting clusters correspond to meaningful objects or regions in the image. 19 refs., 3 figs.
- Research Organization:
- Lawrence Livermore National Lab., CA (USA)
- Sponsoring Organization:
- DOE; USDOE, Washington, DC (USA)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 5903298
- Report Number(s):
- UCRL-JC-106451; CONF-910779--4; ON: DE91008061
- Country of Publication:
- United States
- Language:
- English
Similar Records
Texture analysis improves level set segmentation of the anterior abdominal wall
Gabor transforms and neural networks for automatic target recognition