Name
Abstract | ||
|---|---|---|
One of the most important tasks of an image analysis system is image segmentation, the identification of homogeneous regions in an image. In the literature several methods for segmentation are distinguished. Common are edge detection, split and merge, region growingand clustering techniques. Most of the extensive research on image segmentation in the last three decades has been done for gray scale images. However, as the technical equipment for color image acquisition becomes cheaper and more common, color image analysis becomes more and more important. Nearly all techniques for gray scale image segmentation have been transferred to color images. A survey on color image segmentation can be found in (SK94). Most papers on color segmentation follow the clustering method. Here the pixels are mapped to feature vectors in a feature space. Now statistical methods are appl ied to find some clusters in this feature space. These clusters, re-mapped to the image, form the color segments. A well-known clustering technique is recursive histogram splitting ((Oht85)), applied by many r esearchers ((Cel90), (Tom90)). The advantage of clustering methods is the global view of the data often in form of histograms. However, although histograms provide a global view of the feature data, they do not reflect the spatial information of the underlying image. The extension of clusters in feature space is often ambiguous and the statistical methods trying to solve this problem are computationally expensive. Hanson and Riseman ((HR78)) discuss the advantages and disadvantages of clustering in detail. Region growing techniques start with initial cells, pixels or small regions and let them grow by sequential merging with neighbored, similar regions. The pure local methods tend to chaining mismatches by merging differently colored segments. The centroid-linka ge techniques are sequential methods and are therefore dependent on the choice of starting point and the order in which the pixels are processed (s. (HS85)). We tried to develop a new method combining the advantages of local (simplicity and fastness) and global (robustness and accuracy) techniques. It is a hierarchical region growing method that is inherently paral- lel and therefore independent of the choice of the starting p oint and the order of processing. It uses local and global information and achieves very robust segmentation results in natural color scenes which is also explained by the use of a newly developed color similarity measure. Our idea was first published in (PR93). Since then a lot of improvements have been applied. This paper describes our entire color seg- mentation system, called CSC (Color Structure Code), in detail. In section 2 we introduce the hexagonal, hierarchical island structure on which our method is based. Section 3 describes the actual segmentation method. In Section 4 the new color similarity measure is presented. Section 5 discusses the complexity of our approach. The system is very fast and thus applicable in real world problems. Finally we present some results and conclusions in section 6. |
Year | DOI | Venue |
|---|---|---|
1998 | 10.1007/3-540-63930-6_172 | ACCV (1) |
Keywords | Field | DocType |
natural color scenes,robust segmentation,region growing,color image,edge detection,feature space,spatial information,image analysis,feature vector,image segmentation | Computer vision,Scale-space segmentation,Similarity measure,Pattern recognition,Computer science,Segmentation,Edge detection,Image processing,Image segmentation,Artificial intelligence,Color difference,Color image | Conference |
ISBN | Citations | PageRank |
3-540-63930-6 | 42 | 2.71 |
References | Authors | |
5 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Volker Rehrmann | 1 | 81 | 6.96 |
| Lutz Priese | 2 | 240 | 31.41 |