Some experiments in image segmentation by clustering of local feature values

摘要

This paper describes some attempts to segment textured black and white images by detecting clusters of local feature values and partitioning the feature space so as to separate these clusters. This approach often works well for multispectral images, using a point's spectral signature as its feature vector. The results for grayscale images, using vectors of local feature values, are not quite as good. Factors affecting the performance of this approach are discussed, and methods of improving the results are suggested.