Adaptive Neuro-Fuzzy Inference System (ANFIS) for segmentation of image ROI and Retrieval of ROI based on MP-KDD

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 03 | Mar -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 832
Adaptive Neuro-Fuzzy Inference System (ANFIS) for segmentation of
image ROI and Retrieval of ROI based on MP-KDD
1PROF. HARISH BARAPATRE, 2PRASHANT ROKADE, 3AKSHAY KADAM,
4SACHIN NANAWARE
1234Yadavrao Tasgaonkar Institute Of Engineering And Technology Dept. Of Computer Engineering
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - parallel computing targets problems that are
scalable and possibly distributed, dividing the problem into
smaller pieces. This approach may beexploredtosatisfyreal
time constraints required by augmented reality algorithms.
The implementation is able to provide satisfactory speed up
improvements using CUDA, NVIDIA's architecture for GPU
programming. The aim of this paper is not to present a new
technology, but to show the great improvements that can be
obtained by applying it in computer vision and augmented
reality applications. The MP-KDD algorithm largely reduces
the computational overhead by removing all floating-point
and multiplication operationswhilepreservingthecurrently
popular
SIFT and SURF algorithm essence. The MP-KDD algorithm
can be directly andeffectivelymappedontothepixel-parallel
and row-parallel array processors of the vision chip. The
vision chip architecture is also enhanced to realize direct
memory access (DMA) and random access to array
processors so that the MP-KDD algorithm can be executed
more effectively.
Index Terms- Image retrieval, Objectsegmentation, Object
recognition, Image Databases, Computer Vision.
INTRODUCTION
The digital processors each rasterized by an important
versality and their easy programming. However, in our
approach, a Ewan log processing architecture has been
designed. It high lights a compromise between versality,
parallelism, processing speeds and resolution. The analog
processing operators are fully programmable devices by
dynamic reconfiguration; they can be viewed as a Software-
programmableimage processordedicatedtolow-level image
processing. The main objective will be to design a pixel of
less than 10m, 10 mw it has fill factor of 20%. Thus, with the
increasing scaling of the transistor sin such technology, we
could consider the implementation of more sophisticated
image processing operators dedicated to face localization
and recognition. Other application-specific services such as
those in Earth and Environmental sciencecanbe expectedto
become clearer as projects in these areas mature. is
dominated by the needs of the particle physics community
and here one finds the greatest overlap with existing
European and US efforts. Campus
(computing) Grids can also be expected to grow in
importance. Replica management, access to storage,
scheduling and virtual data are major compute/file Grid
areas. The scale of the particle physics problem emphasizes
the need for robust well-managed grids. The SIMD Pear ray
can efficiently finish low-level pixel-parallel operations, but
it is hard to perform row-parallel and non-parallel
operations. Recently, multi-SIMD vision processor
architecture was proposed, and its FPGA implementation
demonstrated the potential advantage of integrating
different levels of parallel processors into one chip. Second,
previous reported vision chips have one PE along with one
pixel, but the PE’s area is 5–20 times larger than the sensor
pixel’s area. Since the chip area grows quickly with image
size, the image that the chip captured has limited size.
Moreover, this fixed one-to-one PE-pixel mapping
relationship reduces the image processing flexibly. The
amount of result data obtained by the parallel processors is
much smaller than the raw image, and can be fetched by the
MPU through data bus. The MPU runs high-level algorithms
and adjust the sensor parameters on demand. The
instructions for the PE array and the RP array are stored in
an on chip program memory and are issued by a specific
controller, whose operation is controlled by the MPU. Inthis
way, the parallel processors can directly access its program
memory, and the parallel processors can operate in parallel
with the MPU. To achieve higher vision chip performance
with the key point framework, this paper proposes a novel
massively parallel key point detection and description (MP-
KDD) algorithm based on the combination of MP-SIFT
detector and MP-SURF descriptor. The MP-KDD algorithm
largely reduces computational costsbyremovingall floating-
point and multiplication operations while preserving
original SIFT and SURF algorithm essence. The MP-SIFT
detector and the MP-SURF descriptorinthealgorithmcanbe
directly and effectively mapped onto the pixel-parallel and
row-parallel array processors of the vision chip. About600–
760 fps of processing speed can be achieved. To perform the
algorithm more effectively, we alsopropose enhancedvision
chip architecture with support for direct memory access
(DMA) and random access to array processors. The rest of
this paper proceeds as follows. The proposed vision chip
architecture is briefly described. In, after an overview of the
original SIFT and SURF algorithms, the proposed MP-KDD
algorithm is illustrated in detail. The experimental resultsof
the MP-KDD algorithm on a FPGA-based prototype. And
finally, One of the applications is target counting. This
application n counts the e cumulative number of targets
coming in to sight. This can also count only desired targets.
At a frame rate of 0.5 kHz, we found expe rimentally that the

maxi-mum target speed was 20km/h. This counting
technique could be applied, for example, to high-speed in
section of small creatures and crops. Another possible
application is the measurement of high - speed rotation. In
this application, the rotation axis and rotationspeedofa ball
are measured in real-time. As a result of experiments at 1
kHz frame rate, the maximum rotation speed of the object
was 1200rpm. We believe that such rotation measurement
could find applications in ballgames, such as soccer and
baseball. A fast random access to their neighborhood is
compulsory in high-speed feature extraction. In our chip
architecture, one RP can randomly access (onlyread)anyPE
in the same row by column index addressing. Likewise, the
MPU and DMA controller can randomly access (read and
write) any RP by row index addressing. To further enhance
the random access ability, a technique called Delegated
access is proposed in this paper.
EXISTING SYSTEM
In Previous segmentation methods, the processappliesonly
for segmenting an object and that too only for single object.
Most of the methods applies only for binary or gray
images.In existing retrieval system, various algorithms
proposed to improve the Retrieval properties between two
images. They are such as BoW (Bag of words), GIST
detectors, MSER detectors,GMMandSIFTareusedtoextract
the feature. Among this, mixture models are commonlyused
in all detectors. Each model improves at least any of the
parameter which improves the matching property.Pixels
from the segmented object is known as feature pixels.
Depend on features of pixels, the pixel matching performed
between multiple images and then the image retrieved .The
data sets used are commonly availed Real world datasets
such as Oxford buildings and INRIA holidays.
A Maximally Stable Extremal Region (MSER) is a connected
component of an appropriately thre holdedimage.Theword
‘extremal’ refers to the property that all pixels inside the
MSER have either higher (bright extremal regions) or lower
(dark extremalregions) intensity than all the pixels on its
outer boundary. The ‘maximally stable’ in MSER describes
the property optimized in the thres hold selection
process.The set of extremal regions i.e., the set of all
connected components obtained by thres holding, has a
number of desirable properties. Firstly, a monotonic change
of image intensities leaves Eunchanged, since it depends
only on the ordering of pixel intensities which is preserved
under monotonic transformation.Thisensuresthatcommon
photo metric changes modelled locally as linear or affine
leave Eunaffected, even if the camera is non-linear (gamma-
corrected). Secondly, continuousgeometric transformations
preserve topology–pixels from a single connected
component are transformed to a single connected
component.
DISADVANTAGES.
• Most of the process concentrates either
Segmentation or Retrieval.
• Here the image can be described as the histogram
which is used to compute a similarity between the
pair of images.
• By using K-mean clustering algorithm the
vocabularies are build. But its results usually scale
poorly with the size of the vocabulary.
LITERATURE SURVEY
“A CMOS Imager With a Programmable Bit-
SerialColumn-ParallelSIMD/MIMDProcessor”.
Author Name:- Hirofumi Yamashita, and Charles G. Sodini,
The processor is physically arrangedasa denselypacked2-D
processing element (PE) arrayatanimagercolumnlevel.The
digital processor has a multiple-instruction–multiple-
data(MIMD) architecture configuring multiple column-
parallel single-instruction–multiple-data (SIMD)processors.
The prototype im-ager chip with 128×128 pixelsand4×128
PE array designedwith 0.6-µ m technology was fabricated,
and its functionalitywas tested. The estimation of
performance level of the proposedprocessor architecture
with an advanced technology such as the0.09-µ m process
technology showsthattheproposedimagerchiparchitecture
has a potential of giga sum operations per second persquare
millimeter class processing performance.
The proposed image process-ing architecture uses the
system partitioning approach, wherecomputationally
intensive pixel-rate processing alone is imple-mented on an
imager chip to target low-level image process-ing tasks. The
processor is physically arranged as a denselypacked 2-D
processing element (PE) arrayatanimagercolumnlevel.The
digital processor has a multiple-instruction–multiple-data
(MIMD) architecture configuring multiple column-
parallelsingle-instruction–multiple-data (SIMD)processors.
The fullyprogrammable image processor incorporatedin an
imager chipis advantageous over the alternative special-
purpose on-imagerdigital signal processor (DSP) solutions
because of its flexi-bility for an image processing algorithms
The layout pitch of all the datapaths is matched to the pixel
column pitch in order to minimizesilicon area. Image data
flow from the pixel array to the outputregister along the
column direction; thus,noextra wiringforlong-distancedata
transfer is required. The proposed densearray of column-
parallel data path architecture realizes high computational
power with a minimal silicon area.The performance level
ofthe proposed processing architecture with advanced
process technology shows that the proposed imager chip
architecture isone of the candidates for an imager with a
fully programmable high-speed processing performancefor
a variety of consumer imagingapplications.Theaspectof the
imager chip implementing the proposed SIMD/MIMD
processor is shown in As described inSection IV, by
employing the proposed processing scheme, the total
processor silicon area becomes scalable with an
increasingnumber of pixels.Theratiooftheprocessorsilicon
area tothe pixel-array silicon area remains as the number of
pixels increases, if the processing task is the same.
Advantage:-

The advantage of column-parallel processing architecture is
compares three types of on-imagerprocessor architecture
The advantage of a bit-serial architecture is its reducedhardware
complexity.
Disadvantage:-
The bit-parallel processor architecture used in the previous
work makes it difficult to implement a per-column PE array
structure since a multibitbus architecture covers considerable PE
column pitch
“ Parallel Hardware Architecture for Scale
andRotation Invariant Feature Detection”.
Author Name:Vanderlei Bonato, Eduardo Marques, and
George A. Constantinides,The work also proposes specific
hardware optimizations considered fundamental to embed
such a robotic control system on-a-chip.The proposed
architecture is completely stand-alone; it reads the input
data directly from a CMOS image sensor and provides the
results via a Field-ProgrammableGateArray(FPGA)coupled
toan embedded processor. The results may either be used
directly in an on-chip application or accessed through an
Ethernet connection. The system isabletodetectfeaturesup
to 30 frames per second (320£240 pixels) and has accuracy
similar to a PC-based implementation.The main
contributions of this work are: as far we know, the first
complete presentation of an on-chip architecture for the
SIFT algorithm,² the definition of a systemconfigurationand
algorithm modifications appropriate for hardware-
orientated imple-mentation, and² an analysis of the through
put requirement for each SIFTalgorithm function in a real-
time embedded setting. Additionally, this work presents an
on-chip hardware/software co-design, whichgivesflexibility
for the users to customize features descriptors according to
the application needs.
The parallelism requirement of our system is defined based
on the number of operations executed per second in real
time (30 frames p/s),which is computed by multiplying the
system through put with the number of operations
performed to generate a result; by assuming that each
operation takes one clock cycle, which happens when all
fractional circuits are fully pipelined and there are no
pipeline stalls1; and by predicting what is the clock
frequency supported by the target device (FPGA)2wherethe
operations are implemented Although the current
performance satisfies our feature-basedmapping problem,
the software used to associate the feature descriptors has
the most critical time in the case of need ing higher number
of features being extracted per frame. Thesimplestsolutions
to solve this problem in software are to upgrade the NIOS II
softcore processor from a standard to a fast version or to
adopt an FPGA with hardcore processoras generally it has
better performance than a softcore. On the other hand, the
hardware blocks were developed to supportany number of
features per frame. The only parameter that needs to be
adjusted is the internal buffer size between Do Gand Kp
blocks so as to avoid overflow.
Advantage:-
The advantageoftheGaussiankernel symmetryandsave
two multipliers by reusing data from the multiplication
result atk 1 and k2.
One advantage of this dependencyisthe possibilityofhaving
a module slower than the period between two pixels, since
this difference can be compensated when a pixel is rejected
bysome previous function
Disadvantage:-
Our application this is not considered a negative result
since the proportion between false positive and true
matchinghas stayed approximately the same.
PROPOSED SYSTEM
Our work proposes a novel model that concurrently tackles
the problems ofimagesegmentation(ROIsegmentation)and
Retrieval. This kind of retrieval after segmentation provides
matching accuracy between images. This paper studies the
application of the Adaptive Neuro-Fuzzy Inference System
(ANFIS) for segmentation of images (oxford building
dataset). Probabilistic generative model is used for the
problem of image retrieval based on MP-KDD detectors. The
proposed method has the ability to work more than one ROI
in the query image. The fuzzy inference system that we have
considered is a model that maps–input characteristics to
input membership functions, input membership function to
rules a set of output characteristics to output membership
functions, and the function to a single-valued output, or a
decision associated with the output. A network-type
structure similar to that of a neural network, which maps
inputs through input membership functions and associated
parameters, and then through output membershipfunctions
andassociated parameters to outputs, can be used to
interpret the input/output map. The parameters associated
with the membership functions will change through the
learning process. The computation of these parameters (or
their adjustment) is facilitated by a gradient vector, which
provides a measure of how well thefuzzyinferencesystemis
modeling the input/output data foragivensetofparameters.
Once the gradient vector is obtained, any of several
optimization routines could be applied in order to adjustthe
parameters so as to reduce some error measure (usually
defined by the sum of the squared differencebetweenactual
and desired outputs). Anfis useseither back propagationora
combination of least squares estimation and back
propagation for membership functionparameterestimation.
When checking data is presented toanfisas well as training
data, the FIS model is selectedtohave parametersassociated
with the minimum checking data model error.One problem
with model validation formodelsconstructedusingadaptive
techniques is selecting a data set that is both representative
of the data the trained model is intended to emulate, yet
sufficiently distinct from the training data set so as not to
render the validation process trivial.If you have collected a
large amount of data, hopefully this data contains all the
necessary representativefeatures,sotheprocessofselecting
a data set for checking or testing purposes is made easier

ADVANTAGES:-
• The pixel similarity detection can improve than the
previous designs.
• Hence the effectiveness of ROI segmentation and
retrieval, can improve the matching property.
• This comparison determines that our work can offer
better accuracy and higher frame rate within much
less hardware assets than thestate-of-the-artvision
chip in complex tenders. Authors and Affiliation
METHODOLOGY
SOFTWARQE ARCHITECTURE
IMPLIMENTATION
MODULES DESCRIPTION
Pre processing – Filtering – Median filter :-
Initially the input images are preprocessed ,in order to
improve the quality of the images we normallyemploysome
filtering operations. Median filter is used for filtering. The
median filter considers each pixel in the image in turn and
looks at its nearby neighbors to decide whether or not it is
representative of its surroundings. Instead of simply
replacing the pixel value with the median of neighboring
pixel values.The median is calculated by first sorting all the
pixel values from the surrounding neighborhood into
numerical order and then replacing the pixel being
considered with the middle pixel value.
ROI segmentation – ANFIS:-
Adaptive Neuro-Fuzzy system is a kind of artificial neural
network ANFIS used for automatic multilevel image
segmentation.This system consists of multilayer perceptron
(MLP) like network that performs color imagesegmentation
using multilevel thresholding.Thresholdvaluesfordetecting
clusters are found automaticallyusingFuzzyC-means(FCM)
clustering technique. Neural network isemployedtofind the
number of objects automatically from an image. A class of
adaptive networks that arefunctionally equivalent to fuzzy
inferencesystems.ANFIS architectures representing both
theSugeno and Tsukamoto fuzzy models. ANFIS systemuses
two neural network and fuzzy logic approaches. Whenthese
two systems are combined, theymay qualitatively and
quantitatively achieve an appropriateresultthatwill include
either fuzzy intellect orcalculative abilities of neural
network. As other fuzzy systems, the ANFIS structure is
organized of two introductoryand concluding parts which
are linked together by a set of rules. We may recognize five
distinct layers in thestructure of ANFIS network which
makes it as a multi-layer network. A kind of this network,
which is a Sugenotype fuzzy system with two inputsandone
output.
Object Retrieval – MPKDD:-
A massively parallel key pointdetectionanddescription.The
MP-KDD algorithm largely reduces the computational
overhead by removing all floating-point and multiplication
operations while preserving the currently popular SIFT and
SURF algorithm essence.We can freely choose any
combination of detector and descriptor from the SIFT and
SURF algorithmsMassively parallel key point detection and
description (MP-KDD) algorithm is based on the
combination of MP-SIFT detector and MP-SURF
descriptor.To validate ORB, we perform experiments that
test theproperties of ORB relative to SIFT and SURF, for
bothraw matching ability, and performance in image-
matching applications. We also illustrate the efficiency of
ORB by implementing a patch-tracking application on a
smartphone. An additional benefit of ORB is that it is free
fromthe licensing restrictions of SIFT and SURF
RESULT
Fig -1 Image for ROI segmentation
Select image from Database1
Fig -2 Seed selection
First it needs to find regions of interest on the basis of the
brightest pixels having the maximal values because they
represent the most significant regions in theimage. Oncethe
regions of interest are determined, the centroid of each
region needs to found. The resulting centroid pixel is the
seeds for region growing algorithm.

Fig -3 Segmented ROI
By using fuzzy rule Segmented ROI is generated.
Fig -4 Select image for Retrieval
Select image from Database2
Fig -5 Retrieval
Results of the matching process between a query image and
two images of the reference database. As can be observed,
the thresholds used were very conservative so that either
correctly (left) or wrongly (right)retrievedimagesexhibited
a relevant number of potential matches. Subsequently, the
proposed generative model is in charge of filtering out false
matches and providing a refined image ranking.
Fig -6 Retrieved Image
After matching all point, we got true image.
Table- 1
Chart -1 PSNR and MSE
After getting true image and by calculating PSNR and MSE
we get final Report..
SOFTWARE DESCRIPTION
MATLAB® is a high-level technical computing language and
interactive environment for algorithm development, data
visualization, data analysis, and numerical computation.
Using MATLAB, you can solve technical computingproblems
faster than with traditional programming languages,suchas
C, C++, and FORTRAN. Mat lab is a data analysis and
visualization tool which has been designed with powerful
support for matrices and matrix operations. As well as this,
Mat lab has excellent graphics capabilities, and its own
powerful programming language. One of the reasons that
Mat lab has become such an important tool is through the
use of sets of Mat lab programs designed to support a
particular task. These sets of programs are called toolboxes,
and the particular toolbox of interest to us is the image
processing toolbox. Rather than give a description of all of
Mat lab’s capabilities, we shall restrictourselvestojustthose
aspects concerned with handling of images. We shall
introduce functions, commands and techniques as required.
A Mat lab function is a keyword which accepts various
parameters, and produces somesortofoutput:for examplea
matrix, a string, a graph. Examples of such functions are sin,
I’m read, and I’m close. There are many functions in Mat lab,
and as we shall see, it is very easy (and sometimes
necessary) to write our own .Mat lab’s standard data type is

the matrix all data are considered to be matrices of some
sort. Images, of course, are matrices whose elements are the
grey values (or possibly the RGB values) of its pixels. Single
values are considered by Mat lab to be matrices, while a
string is merely a matrix of characters; being the string's
length. In this chapter we will look at the more generic Mat
lab commands, and discussimagesinfurtherchapters.When
you start up Mat lab, you have a blank window called the
Command Window_ in which you enter commands. Given
the vast number of Mat lab’s functions, and the different
parameters they can take, a command line style interface is
in fact much more efficient than a complex sequence of pull-
down menus. You can use MATLAB in a wide range of
applications, including signal and image processing,
communications, control design, test and measurement
financial modeling and analysis. Add-on toolboxes
(collections of special-purpose MATLAB functions) extend
the MATLAB environment to solve particular classes of
problems in these application areas.MATLAB provides a
number of features for documenting and sharing yourwork.
You can integrate your MATLAB code with other languages
and applications, and distribute your MATLAB algorithms
and applications.
When working with images in Mat lab,therearemanythings
to keep in mind such as loading an image, using the right
format, saving the data as different data types, how to
display an image, conversion between different image
formats. Image Processing Toolbox provides a
comprehensive set of reference-standard algorithms and
graphical tools for image processing, analysis, visualization,
and algorithm development. You can perform image
enhancement, image de blurring, feature detection, noise
reduction, image segmentation, spatial transformations,and
image registration. Many functions in the toolbox are
multithreaded to take advantage of multicore and
multiprocessor computers.
Key Features
•High-level language for technical computing
•Development environment for managing code, files, and
data
•Interactive tools for iterative exploration, design, and
problem solving
•Mathematical functionsforlinearalgebra,statistics,Fourier
analysis, filtering, optimization, and numerical integration
•2-D and 3-D graphics functions for visualizing data
•Tools for building custom graphical user interfaces
•Functions for integrating MATLAB based algorithms with
external applications and languages, such as C, C++, Fortran,
Java, COM, and Microsoft Excel.
FUTURE SCOPE
In this paper we have proposed a generative probabilistic
model that concurrently tackles image retrieval and roi
segmentation problems. By jointly modeling several
properties of true matches, namely: objects undergoing a
geometric transformation, typical spatial location of the
region of interest, and visual similarity, our approach
improves the reliability of detected true matches between
any pair of images. Furthermore, the proposed method
associates the true matches with any of the considered
foreground components in the image and assigns the rest of
the matches to a background region, what allows it to
perform suitable roi segmentation. We have conducted a
comprehensive assessment of the proposed method. Our
results on two well-known databases, oxford building and
holidays, prove that it is highly competitive in traditional
image retrieval tasks, providing favorable results in
comparison to most of the state-of-the-art systems.
Regarding roi segmentation, assessed on the oxford
database, the proposed model outperformed ransac, the
most well-known geometric approach. In our opinion these
results are due to two main reasons: first, our model jointly
manages several properties of true matches; and second, by
considering the whole set of reference images at once, the
proposed method provides a robust method for estimating
the actual geometric transformation undergone by the
objects. In particular, by computingtheposteriorprobability
that a match is considered as true (e.g. It belongs to any of
the considered foregroundcomponents),successfullyrejects
outliers in the estimation of the geometric transformation.
This out lier rejection ability notably improves when all the
reference images are jointly considered in comparison to
traditional techniques where each pair of images(queryand
reference) is addressed independently. In addition, our
model can also work in scenarios where there is more than
one object-of-interest in the query image. To assess the
performance of the proposed model, we have conducted
three different experiments: a multi-class category
segmentation experiment on the ethz toys dataset; a multi
object detection experiment on the rgb-d dataset; and a
multiview object retrieval experimentontheoxfordbuilding
dataset. For the first two cases, we developed and tested a
method for automatically selecting the number k of objects
of-interest in the query image, with results very close to
those ones achieved with the optimal k in each case. In the
third experiment, the results showed a significant
performance improvement when the number of foreground
objects considered by the model fitted the actual number of
objects-of-interest. These results allow us to conclude that
the performance of the retrieval process can be notable
improved when different views of the object-of-interest are
available.
CONCLUSION
This paper proposes a scale-invariant keypoints
detection and feature description algorithm named as MP-
KDD for vision chips. The MP-KDD algorithm is inspired by
the combination of the original SIFTdetector and SURF
descriptor, but involves only simple fixed-point operations
without any multiplicationrequirement. The MP-KDD
algorithm can be directly and efficiently mapped onto the
massively parallelvision chip architecture. To be better
compatible with the MP-KDD algorithm, this paper also

enhancesthe vision chip architecture with multiple levels of
array random access.
The visual measurements basedonmulti-targettracking,
target counting and rotation measurement. The
experimentalresultsusingthevisionchip wehavedeveloped
show that both methods excel in the points, high-precision,
high-frame-rate observation and flexibility. These visual
measurements, providing such advantages, are expected to
be applied in various fields, such as in spection, industrial
ap-plications, sport measurement, robot control, and so on.
REFERENCES
[1] Ishikawa M, Ogawa K, KomoroT, etal. A CMOS vision chip
with SIMD processing element array for1 msimagepro-
cessing. In: Proceedings of IEEE Solid-State Circuits
Conference Digest of Technical Papers (ISSCC), San
Francisco, 1999. 206–207
[2] Shi C, Yang J, Han Y, et al. A 1000 fps vision chip based on
a dynamically reconfigurable hybrid architecture
comprising a PE array and self-organizing map neural
network. In: Proceedings of IEEE Solid-State Circuits
Conference Digest of Technical Papers (ISSCC), San
Francisco, 2014. 128–129
[3] Komoro T, Ishii I, Ishikawa M, et al. A digital vision chip
specialized for high-speed target tracking. IEEE Trans
Electron Dev, 2003, 50: 191–199
[4] Dudek P, Hicks P J.A general-purposeprocessor-per-pixel
analog SIMD vision chip. IEEE Trans Circuits Syst I-
RegulPap, 2005, 52: 13–20
[5] Ishii I, Yamamoto K, Kubozono M. Higher order
autocorrelation vision chip. IEEE Trans Electron Dev,
2006, 53:1797–1804
[6] Miao W, Lin Q Y, Zhang W C, etal. A programmable SIMD
vision chip for real-time vision applications. IEEE J Solid-
State Circuits, 2008, 43: 1470–1479
[7] Dubois J, Ginhac D, Paindavoine M, et al. A 10 000 fps
CMOS sensor with massively parallel image processing.
IEEEJ Solid-State Circuits, 2008, 43: 706–717
[8] Yamashita H, Sodini C.A CMOS imager with a
programmable bit-serial column-parallel SIMD/MIMD
processor. IEEE Trans Electron Dev., 2009, 56: 2534–
2545
[9] Lin Q Y, Miao W, Zhang W C.A 1000 frame/s
programmable vision chip with variable resolution and
row-pixel-mixed parallel image processors. Sensors,
2009, 9: 5933–5951
[10] Cheng C C , Lin C H, Li C T, etal. I Visual: an
intelligent visual sensor SoC with 2790 fps image sensor
and 205
[11] Zhang W C, Fu Q Y, Wu N J.A programmable visionchip
based on multiple levels of parallel processors. IEEE J
Solid-State Circuits, 2011, 46: 2132–2147
[12] Lopich A, Dudek P.A SIMD cellular processor array
vision chip with asynchronous processing capabilities.
IEEE Trans Circuits Syst I-Regul Pap, 2011, 58: 2420–
2431
[13] Cottini N, Gottardi M, Massari N, et al. A 33µW 64×64
pixel vision sensor embedding robust dynamic
background subtraction for event detection and scene
interpretation. IEEE J Solid-State Circuits, 2013, 48:
850–86314 Gauglitz S, H¨ollerer T, Turk M. Evaluation
of interest point detectors and feature descriptors for
visual tracking. Int J Compute Vis, 2011, 94: 335–360
[14] Lowe D G. Distinctive image features from scale-
invariant key points. Int J Compute Vis, 2004, 60: 91–
110
[15] Bay H, Ess A, Tuytelaars T, et al. Speeded-up robust
features (SURF). Compute Vis Image Understate,2008,
110:346–359
[16] Bonato V, Marques E, Constantinides G A.A parallel
hardware architecture for scale and rotation invariant
feature detection. IEEE Trans Circuits Syst Video
Technol, 2008, 18: 1703–1712
[17] Kim K, Lee S, Kim J Y, et al.A configurable
heterogeneous multicore architecture with cellular
neural network forreal-time object recognition. IEEE
Trans Circuits Syst Video Technol, 2009, 19: 1612–
1622
[18] Zhang W L, Liu L B, Yin S Y, etal. An efficient VLSI
architecture of speeded-up robust feature extraction
for high resolution and high frame rate video. SciChina
Inf Sci, 2013, 56: 072402
[19] Zhou Y F, Cao Z X, Qin Q, et al. A 1000 FPS high speed
CMOS image sensor with low noise global shutter
pixels. Sci China Inf Sci, 2014, 57: 042405
[20] C surka C, Dance C R, Fan L, et al. Visual categorization
with bags of key points. In: Proceedings of ECCV
International Workshop on Statistical Learning in
Computer Vision, Prague, 2004. 1–7

Adaptive Neuro-Fuzzy Inference System (ANFIS) for segmentation of image ROI and Retrieval of ROI based on MP-KDD

Recommended

More Related Content

What's hot (16)

Similar to Adaptive Neuro-Fuzzy Inference System (ANFIS) for segmentation of image ROI and Retrieval of ROI based on MP-KDD (20)

More from IRJET Journal (20)

Recently uploaded (20)

Adaptive Neuro-Fuzzy Inference System (ANFIS) for segmentation of image ROI and Retrieval of ROI based on MP-KDD