Professional Documents
Culture Documents
Abstract—Mathematical morphology is a promising computer them employ a pipeline technique in which a raster-scan image
paradigm based on set theory and has many applications in is sequentially fed into a processing element (PE) array and the
image processing. Although some architectures have been pro- morphological operations are carried out in parallel in each PE.
posed, there are as yet no compact, practical computers that can
handle a variety of morphological operations with large, complex Since the functions of the PEs and the network structure are fully
structuring elements at video rates. This has prevented the great tuned to morphology, other operations crucial to practical image
potential of morphology from being fully realized. This paper processing can not be performed. The fixed network structure
describes a morphology processing method that uses a highly also limits the size and shape of the structuring elements. Fur-
parallel two-dimensional (2-D) cellular automaton architecture thermore, there are at most several dozen PEs. This prevents the
called it CAM2 (Cellular AutoMata on Content Addressable
Memory). New mapping methods achieve high-throughput com- full use of the abundant parallelism (pixel order) of morphology,
plex morphology processing. Evaluation results show that CAM2 and, as a result, the processing speed of the pipeline type is not
performs one morphological operation for basic structuring high enough for many real-time applications. Against this back-
elements within 30 s. Furthermore, CAM2 can also handle an drop, it is pretty clear that none of these conventional architec-
extremely large and complex structuring element of 100 100 at tures are suitable for building a morphology processing platform
video rates. CAM2 will increase the potential use of morphology
and make a significant contribution to the development of various that satisfies the above three prerequisites.
real-time image processing systems. A two-dimensional (2-D) cellular array architecture [14]–[16],
which consists of 2-D PEs and interconnection networks, is an-
Index Terms—Cellular automaton, content addressable
memory, mathematical morphology, pattern spectrum, real-time other candidate for the platform because it is the most natural
image processing. architecture for morphology. The drawback of the conventional
fully parallel approach is the huge amount of hardware involved.
At most, only several dozen PEs can be embedded onto one
I. INTRODUCTION VLSI chip. So, enormous numbers of VLSI chips are required to
realize pixel-order parallelism, which is crucial for extracting the
M ATHEMATICAL morphology [1] is an image trans-
formation technique that locally modifies geometric
features through set operations. It is a powerful tool with various
performance. Moreover, 2-D interconnection networks cause I/O
bottlenecks, so it is difficult to increase the number of PEs even if
applications [2]–[6], such as nonlinear image filtering, noise sup- state-of-the-art LSI technology is used.
pression, smoothing and shape recognition; and it is becoming This paper describes a morphology processing method that
very common in image processing. There are three prerequisites uses CAM . CAM is a compact, high-performance, flexible,
for the fuller realization of the potential of morphology: and highly parallel 2-D cellular automata (CA) [17]. CAM can
• complex processing combining various morphological op- attain pixel-order parallelism on a single PC board because it
erations (including other operations, such as discrete-time is composed of a content addressable memory (CAM), which
cellular neural networks [7], linear filtering [8], and area makes it possible to embed great numbers of PEs, corresponding
calculation); to CA cells, onto one VLSI chip. New mapping methods achieve
• processing with large and complex structuring elements; high-throughput complex morphology processing. Evaluation
• high-speed (real-time) processing. results show that CAM performs one morphological operation
for basic structuring elements within 30 s. This means that
The achievement of these goals requires hardware with ex-
more than 1000 operations can be carried out on the whole
tremely high performance and high-frequency memory accesses.
pixel image at video rates (33 ms). CAM also handles an
That makes general-purpose sequential machines like personal
extremely large and complex structuring element
computers (PC) and workstations (WS) totally unsuitable.
at video rates. Furthermore, it performs practical image pro-
To address these problems, some special-purpose architec-
cessing, such as pattern spectrum and multiple object tracking,
tures for morphology have been proposed [9]–[13]. Most of
through a combination of morphology and other algorithms.
Section II presents the features of CAM . This is followed by
a description of the morphology processing method including
Manuscript received February 26, 1998; revised June 12, 2000. This work was
supported by Y. Sakai, O. Karatsu, K. Takeya, and R. Kasai. The associate editor pattern spectrum processing in Sections III and IV. After dis-
coordinating the review of this manuscript and approving it for publication was cussing the application development environment in Section V,
Dr. Sridhar Lakshmanan. performance evaluation results and some examples of image
The authors are with NTT Lifestyle and Environmental technology Labora-
tories, Kanagawa 243-0198, Japan (e-mail: ikenaga@aecl.ntt.co.jp). processing combining morphology and other algorithms are
Publisher Item Identifier S 1057-7149(00)10069-7. presented in Section VI.
1057–7149/00$10.00 © 2000 IEEE
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
IKENAGA AND OGURA: REAL-TIME MORPHOLOGY PROCESSING 2019
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
2020 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 12, DECEMBER 2000
SP
(1)
(2)
FSP
(3)
(4)
FP
(5)
(6)
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
IKENAGA AND OGURA: REAL-TIME MORPHOLOGY PROCESSING 2021
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
2022 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 12, DECEMBER 2000
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
IKENAGA AND OGURA: REAL-TIME MORPHOLOGY PROCESSING 2023
TABLE I
PROCESSING TIME FOR BASIC SEs (s)
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
2024 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 12, DECEMBER 2000
Fig. 12. Pattern spectrum for images with and without crack.
TABLE II
PATTERN SPECTRUM PROCESSING TIME
only takes 1.6 ms even for a relatively large image ( ), Fig. 14. Example of CAM PL.
more than 30 ms is available in which to perform morphology
or other CA algorithms for real-time, or video rate (33 ms), ap- tively. Thus, a pattern spectrum can also be obtained at video
plications. rates.
2) Pattern Spectrum: Fig. 12 shows examples of pattern 3) Multiple Object Tracking: Another example of image
spectrum processing (SE: circle) for two different images, processing using CAM is shown in Fig. 13. By applying
in which one object has a crack and the other one does not. various morphological operations to perform line erasure, edge
As shown in Fig. 12, since the size and shape of the objects detection, hole filling and noise reduction, a binary image of
in image 1 are uniform, the spectrum concentrates on scales 5 target objects can be obtained. The processing requires 15
and 6. In contrast, since the object in image 2 has cracks, the morphological operations with various structuring elements.
spectrum is scattered. Since the features of the spectra are quite CAM can do it in just 200 s. As shown in Section VI-B1,
different, it is easy to distinguish them. the data loading and retrieval times are 0.1 ms and 1.6 ms for
Table II shows the processing time for the images. As shown and images, respectively. The processing
in the table, the processing time for opening increases with the can be finished at video rates.
scale because the size of the structuring element becomes larger Fig. 14 shows an example of CAM PL [25] for the pro-
as the scale increases. In contrast, the processing time for the cessing in Fig. 13. Here, “copy_8” means the intra-word copy
area calculation is constant, and is about that of the con- of 8 bits and “sub_data8” means pixel-by-pixel subtraction of 8
ventional method using the data I/O and the external processor. bits. “Dilation” and “erosion” are the dedicated operations for
It takes only 0.6 s per scale for the rest of the processing, such describing morphological operations with various structuring
as pixel-by-pixel subtaction. elements, as mentioned before. Using these operations, the pro-
It takes about 1 ms for the whole pattern spectrum processing. cessing is described in only 20 operations.
When data loading time is included, the total time becomes As discussed above, CAM efficiently performs not only
1.1 ms and 1.8 ms for and images, respec- morphology, but also other CA-based algorithms. Using these
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
IKENAGA AND OGURA: REAL-TIME MORPHOLOGY PROCESSING 2025
ACKNOWLEDGMENT
The authors would like to thank Y. Takahashi, Y. Fujino, T.
Fig. 16. CAM board with 256 2 256 CA cells. Tsuchiya, T. Nakanishi, M. Nakanishi, and E. Hosoya for their
many valuable suggestions and constructive discussions.
algorithms, the center points of target objects and a distance
map for them can be obtained as shown in Fig. 15. By applying REFERENCES
the processings in Figs. 13 and 15 to the input image and by [1] J. Serra, Image Analysis and Mathematical Morphology. New York:
finding the center points nearest those in the previous frame, Academic, 1982.
[2] P. Maragos, “Tutorial on advances in morphological image processing
multiple object tracking can be performed. and analysis,” Opt. Eng., vol. 26, 1987.
These examples demonstrate that CAM is flexible enough to [3] R. M. Haralick, S. R. Sternberg, and X. Zhuang, “Image analysis using
perform practical image processing employing a combination of mathematical morphology,” IEEE Trans. Pattern Anal. Machine Intell.,
vol. 9, no. 4, pp. 532–550, 1987.
morphology and other algorithms. [4] L. Vincent, “Graphs and mathematical morphology,” Signal Process.,
vol. 16, no. 4, pp. 365–388, 1989.
[5] S. Yamamoto, M. Matsumoto, Y. Tateno, T. Iinuma, and T. Matsumoto,
VII. CONCLUSION “Quoit filter—A new filter based on mathematical morphology to ex-
tract the isolated shadow, and its application to automatic detection of
This paper has described a morphology processing method lung cancer in X-ray CT,” in Proc. 13th Int. Conf. Pattern Recognition
based on a highly-parallel 2-D cellular automata called CAM (ICPR’96), vol. 2, 1996, pp. 3–7.
and has presented some evaluation results. New mapping [6] Y. Takahashi, A. Shio, and K. Ishii, “Morphology based thresholding for
character extraction,” IEICE Trans. Inform. Syst., vol. E76-D, no. 10, pp.
methods using maskable search, partial & parallel write and 1208–1215, 1997.
hit-flag shift achieve high-throughput complex morphology [7] H. Harrer, “Multiple layer discrete-time cellular neural networks using
processing. Evaluation results show that CAM performs one time-variant templates,” IEEE Trans. Circuits Syst. II, vol. 40, pp.
191–199, Mar. 1993.
morphological operation for basic structuring elements within [8] E. R. Dougherty and P. A. Laplante, Real-Time Imaging. New York:
30 s. This means that more than 1000 operations can be IEEE Press, 1995.
carried out on an entire pixel image at video rates (33 [9] M. Hassoun, T. Meyer, P. Siqueira, J. Basart, and S. Gopalratnam, “A
VLSI gray-scale morphology processor for real-time NDE image pro-
ms). CAM can also handle an extremely large and complex cessing applications,” SPIE Image Algebra Morphological Image Pro-
structuring element at video rates. Furthermore, cessing, 1990.
CAM can perform practical image processing, such as pattern [10] R. Lin and E. K. Wong, “Logic gate implementation for gray-scale mor-
phology,” Pattern Recognit. Lett., vol. 13, no. 7, 1992.
spectrum and multiple object tracking, through a combination [11] C. H. Chen and D. L. Yang, “Realization of morphological operations,”
of morphology and other algorithms. Thus, CAM will enable IEE Proc. Circuits Devices Systems, vol. 142, 1995.
fuller realization of the potential of morphology and make a [12] L. Lucke and C. Chakrabarti, “A digital-serial architecture for gray-
scale morphological filtering,” IEEE Trans. Image Processing, vol. 4,
significant contribution to the development of real-time image pp. 387–391, Mar. 1995.
processing systems based on morphology and other algorithms. [13] E. R. Dougherty and D. Sinha, “Computational gray-scale mathemat-
ical morphology on lattices (a comparator-based image algebra)—Part
I: Architecture,” Real-Time Imag., vol. 1, pp. 69–85, 1995.
APPENDIX [14] T. Kondo et al., “Pseudo MIMD array processor-AAP2,” in Proc. 13th
Symp. Computer Architecture Conf., 1986, pp. 330–337.
We have completed our development of a dedicated 1-Mb [15] Thinking Machines Corp, Connection machine model, CM-2 tech. sum-
mary, Ver. 5.1, 1989.
CAM LSI for CAM [29], [30]. We fabricated a chip capable [16] J. R. Nickolls, “The design of the MasPar MP-1: A cost-effective mas-
of operating at 56 MHz and 2.5 V using 0.25- m full-custom sively parallel computer,” in Proc. COMPCON Spring’90, 1990, pp.
CMOS technology with five aluminum layers. Since it has 16k 25–28.
[17] T. Ikenaga and T. Ogura, “CAM : A highly-parallel 2D cellular au-
words, or CA cells, a single chip can process pixels in tomata architecture,” IEEE Trans. Comput., vol. 47, pp. 788–801, July
parallel. We have also developed a prototype CAM board using 1998.
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.
2026 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 12, DECEMBER 2000
[18] T. Ogura, M. Nakanishi, T. Baba, Y. Nakabayashi, and R. Kasai, Takeshi Ikenaga (M’95) received the B.E. and M.E.
“A 336-kbit content addressable memory for highly parallel image degrees in electrical engineering from Waseda Uni-
processing,” in Proc. Custom Integrated Circuits Conf. (CICC’96), versity, Tokyo, Japan, in 1988 and 1990, respectively.
1996, pp. 273–276. He joined LSI Laboratories, Nippon Telegraph
[19] T. Ogura and M. Nakanishi, “CAM-based highly-parallel image and Telephone Corporation (NTT), in 1990, where
processing hardware,” IEICE Trans. Electron., vol. E80-C, no. 7, pp. he has been working on the research of the design
868–874, 1997. and test methodologies for high-performance ASICs.
[20] Y. Fujino, T. Ogura, and T. Tsuchiya, “Facial image tracking system He is presently a Senior Research Engineer with
architecture utilizing real-time labeling,” in Proc. SPIE VCIP’93, 1993. the Parallel Processing Systems Research Group,
[21] M. Nakanishi and T. Ogura, “A real-time CAM-based Hough trans- NTT Lifestyle and Environmental Technology
form algorithm and its performance evaluation,” 13th Int. Conf. Pattern Laboratories, Kanagawa, Japan. His current interests
Recognition (ICPR’96), vol. 2, pp. 516–521, 1996. are highly parallel system design and its applications to computer vision. In
[22] M. Nakanishi and T. Ogura, “Real-time extraction using a highly par- 1999–2000, he was a Visiting Researcher with the University of Massachusetts,
allel Hough transform board,” Proc. IEEE Int. Conf. Image Processing Amherst.
(ICIP’97), vol. 2, pp. 582–585, 1997. Mr. Ikenaga is a member of the Institute of Electronics, Information, and
[23] M. Meribout, M. Nakanishi, and T. Ogura, “Hough transform imple- Communication Engineers (IEICE) of Japan, and the Information Processing
mentation on a reconfigurable highly parallel architecture,” in Proc. Society of Japan (IPSJ). He received the IEICE Research Encouragement Award
Computer Architectures Machine Perception (CAMP’97), 1997, pp. in 1992 for his paper “A test pattern generation for arithmetic execution units.”
276–279.
[24] E. Hosoya, T. Ogura, and M. Nakanishi, “Real-time 3D feature
extraction hardware algorithm with feature point matching capability,”
in Proc. IAPR Workshop Machine Vision Applications (MVA’96), 1996,
pp. 430–433.
[25] T. Ikenaga and T. Ogura, “A DTCNN universal machine based on
highly-parallel 2-D cellular automata CAM ,” IEEE Trans. Circuits Takeshi Ogura (M’86) received the B.S., M.S., and
Syst. I, vol. 45, pp. 538–546, May 1998. Ph.D. degrees in electrical engineering from Osaka
[26] P. Maragos, “Pattern spectrum and multiscale shape representation,” University, Osaka, Japan, in 1976, 1978, and 1991,
IEEE Trans. Pattern. Anal. Machine Intell., vol. 11, pp. 701–716, July respectively.
1989. In 1978, he joined Musashino Electrical Commu-
[27] K. Sudo, J. Yamato, and A. Tomono, “Determining gender using mor- nication Laboratories, Nippon Telegraph and Tele-
phological pattern spectrum” (in Japanese), IEICE Trans. Inform. Syst., phone Public Corporation (NTT), Tokyo, Japan. He
vol. J80-D-II, no. 5, pp. 1037–1045, 1997. is currently an Executive Manager with the Multi-
[28] E. Sternheim et al., Digital Design with Verilog HDL. New York: Au- media Electronics Laboratory of NTT Lifestyle and
tomata, 1990. Environmental Technology Laboratories, Kanagawa,
[29] T. Ikenaga and T. Ogura, “A fully-parallel 1 Mb CAM LSI for real-time Japan. He is engaged in the research and development
pixel-parallel image processing,” in IEEE Int. Solid-State Circuits Conf. of CAM LSIs and their applications and is also engaged in the development of
(ISSCC99) Dig. Tech. Papers, 1999. image processing and encoding LSIs.
[30] T. Ikenaga and T. Ogura, “A fully parallel 1-Mb CAM LSI for real-time Dr. Ogura is a member of the Institute of Electronics, Information, and Com-
pixel-parallel image processing,” IEEE J. Solid State Circuits, vol. 35, munication Engineers (IEICE) of Japan and the Information Processing Society
pp. 536–544, Apr. 2000. of Japan (IPSJ).
Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on July 27,2021 at 15:52:16 UTC from IEEE Xplore. Restrictions apply.