Abstract
In many image analysis applications, objects of rographics. interest exhibit different texture than their surrounding areas. Texture features have been used as the initial step towards elective image segmentation and classification. Among texture features, morphological granulometry texture features are the least utilized partly due to the immense amount of time needed for their computation. This is especially so when structuring elements of any geometry are utilized in obtaining a granulometry. The computational problem is made even greater when large images, such as mammograms are to be analyzed. In addition to high computation time, traditional granulometry computing approaches also require a considerable amount of memory. These factors are the major restraining factors in the use of morphological granulometric features. These features are increasingly being found to be elective in applications such as medical image interpretation. Hence, there is a need to find efficient ways to compute grey-scale granulometry. Some fast approaches are available for computing grey-scale granulometry when flat linear structuring elements are used. However, for many images such as medical images, textures are rarely formed by hat linear sub-structures. This work is an attempt to investigate, and implement efficient methods of computing the granulometry texture features for structuring elements of any geometry. First various approaches for computing granulometry are discussed and compared. Those found efficient are then benchmarked on three machines (Sun SPARC 20, Pentium 11 based Dell 400 workstation, and SGI Power Challenge 10000XL). Parallel processing implementation on two multiprocessor machines (SGI Power Challenge IOOOOXL and Dell 400 workstation) is also performed; followed approach on an advanced DSP processor-TMS320C80.
Patel, Manish J (1998). Implementation of efficient algorithms for the computation of morphological texture features. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1998 -THESIS -P3774.