Simple techniques like thresholding are adequate for high contrast pictures, but they fail whenever the picture has gradual variations in brightness and color or when the different regions are characterized by texture rather than color. In this chapter we describe a general split-and-merge algorithm using the QPT (or pyramid) as its basic data structure. In addition, we outline certain simple editing techniques and how to obtain ordered lists of the boundary points from the output of such an algorithm. The description is in terms of predicates defined over regions rather than specific brightness or color criteria. Therefore, the algorithm is a frame which can be “dressed” according to the problem at hand. Segmentation according to brightness or color is relatively easy and a specific example of the algorithm for this case has been reported by HOROWITZ and PAVLIDIS [5.1]. In this form the algorithm is closely related to the various region growing algorithms described in the previous chapter. However, it can also be used for segmentation according to texture by keeping the same data structure and defining different predicates. At the time of this writing the characterization of texture is very much an open problem. There is, probably, general agreement that the second order statistics of the brightness function describe texture but the computation of many of them is quite time consuming. Furthermore, there is evidence that local shape recognition may be involved as well [5.2]. Interesting comparisons of various types of texture algorithms have been published recently by WESZKA et al. [5.3] and by CONNERS and HARLOW [5.4]. The adaptability of the general algorithm to segmentation by texture has been demonstrated in a preliminary study using the Fourier power spectrum [5.5].
[1]
Theodosios Pavlidis,et al.
The editing of picture segmentations using local analysis of graphs
,
1977,
CACM.
[2]
Béla Julesz,et al.
Visual Pattern Discrimination
,
1962,
IRE Trans. Inf. Theory.
[3]
H. Feng,et al.
The generation of polygonal outlines of objects from gray level pictures
,
1975
.
[4]
Theodosios Pavlidis,et al.
Picture Segmentation by a Tree Traversal Algorithm
,
1976,
JACM.
[5]
Ruzena Bajcsy,et al.
Image filtering-A context dependent process
,
1975
.
[6]
William B. Thompson,et al.
Computer Diagnosis of Pneumoconiosis
,
1974,
IEEE Trans. Syst. Man Cybern..
[7]
Mohamad Adnan Al-Alaoui,et al.
Application of constrained generalized inverse to pattern classification
,
1976,
Pattern Recognit..
[8]
Robert M. Haralick,et al.
Textural Features for Image Classification
,
1973,
IEEE Trans. Syst. Man Cybern..
[9]
Jerome A. Feldman,et al.
Decision Theory and Artificial Intelligence: I. A Semantics-Based Region Analyzer
,
1974,
Artif. Intell..
[10]
Azriel Rosenfeld,et al.
A Comparative Study of Texture Measures for Terrain Classification
,
1975,
IEEE Transactions on Systems, Man, and Cybernetics.