DocumentCode
3179990
Title
The Dynamic Decision Switch for Multiple Pixel Connected Component Labeling Algorithm
Author
Wang, Wen-Shan ; Hung, Ya-Fei ; Yang, Jen-Kuei ; Tseng, Shau-Yin
Author_Institution
SoC Technol. Center, Ind. Technol. Res. Inst., Hsinchu, Taiwan
fYear
2009
fDate
1-3 Dec. 2009
Firstpage
140
Lastpage
145
Abstract
Connected component labeling is an indispensable and one of most time consuming tasks of the applications in computer vision. Many labeling algorithms have been introduced, such as scan plus connection table, scan plus union-find, and contour tracing etc. They would rather use byte data than bit data to represent the binary pixel, which is either 1 or 0, due to the heavy cost of bitwise operations. This paper will propose a mechanism employing bit data to stand for the binary image pixels and labeling multiple pixels in one labeling process so that it can turn the weakness of bit data into the strength. According to the test results run in ARM926EJ-S, this new mechanism can double the speed of the scanning and analysis phases of an array based scan plus union-find algorithm. Besides, the much smaller binary image buffer needed by this mechanism is critical for the limited hardware-resource embedded devices, which are implemented in the field of computer vision gradually.
Keywords
computer vision; decision trees; image resolution; image retrieval; ARM926EJ-S; array based scan plus union-find algorithm; binary image buffer; binary image pixels; computer vision; dynamic decision switch; hardware-resource embedded devices; multiple pixel connected component labeling algorithm; Algorithm design and analysis; Computer industry; Computer vision; Decision trees; Information analysis; Labeling; Phased arrays; Pixel; Switches; Testing;
fLanguage
English
Publisher
ieee
Conference_Titel
Digital Image Computing: Techniques and Applications, 2009. DICTA '09.
Conference_Location
Melbourne, VIC
Print_ISBN
978-1-4244-5297-2
Electronic_ISBN
978-0-7695-3866-2
Type
conf
DOI
10.1109/DICTA.2009.30
Filename
5384991
Link To Document