2. DUTTA: MEDICAL DATA COMPRESSION AND TRANSMISSION IN WANETs 779
Fig. 2. Block diagram of the MDC technique at PHC Station.
principle component analysis is followed by DCT. Vector
quantization is used to code the transformed images. A survey
of compression algorithms for color retinal images is presented
in [8]. The authors in [9] have designed a compressor system
for endoscopic images, which integrates a CMOS sensor with
a JPEG compression engine in a capsule. In [10], endoscopic
images of wireless capsule are compressed using DCT and
Golomb-Rice (GR) coding. None of the aforesaid literature for
medical image compression are robust to transmission errors.
These algorithms are therefore not meaningful for medical
data transmission applications since in practice transmission
channels are not completely noiseless.
Ad-hoc On-Demand Distance Vector Routing (AODV) is
an on-demand routing protocol for WANETs [11]. Nodes do
not depend on active paths neither store any routing infor-
mation nor take part in any periodic routing table exchanges.
Saida and Mellouk in [12] proposed a delay-oriented adaptive
routing protocol called adaptive mean delay routing (AMDR)
by using the reinforcement learning mechanism. In [13], a
distributed routing scheme is designed to offer scalability and
extend the network lifetime in a mobile ad hoc network.
AMDR uses forward exploration packets (FEP) to gather
delay information of available paths that will be used by
backward exploration packets (BEP) to update routing table
entries throughout the network. In [14], a model to estimate
path duration in a MANET using the random way point
mobility model is proposed. A multicast algorithm is pre-
sented in [15] to increase the lifetime of node and network
in the mobile ad hoc network. Two energy-aware routing
algorithms for wireless ad hoc networks, called reliable min-
imum energy cost routing (RMECR) and reliable minimum
energy routing (RMER) are presented in [16] to address
energy-efficiency, reliability, and network lifetime of ad hoc
networks.
The main limitation of the above protocols is that they
concentrate on the discovery of routes satisfying certain quality
of service (QoS) requirements such as minimum hop count
and delay. However, various issues (e.g., network reliability,
low routing delay, long battery lifetime, etc.) that have to take
into account when routing the compressed MDPs. In summary,
there is no work in the literature on routing protocols for
WANETs that can work with compressed MDPs.
Contribution: The main contributions of the paper can be
summarized are as follows:
• First, we propose a novel medical data compression
(MDC) technique that reduces the size of MDPs. The
MDC technique compresses and decompresses MDPs at
PHC station and CC center, respectively. In addition,
robustness is achieved in the presence of transmission
errors. The technique addresses the compression of MDPs
in the form of color images, such as, endoscopic images,
where a robust and efficient compression of color images
is still a challenging problem in the literature [19], [20].
Moreover, the MDC technique also effectively works with
other forms of the medical data, e.g. electrocardiography
images, magnetic resonance imaging, etc.
• Various issues (e.g., network reliability, low routing delay,
and long battery lifetime) have to take into account
when routing the compressed MDPs in WANETs. Next,
we propose a fuzzy-logic based route selection (FRS)
technique that optimizes among of the issues related to
the routing of the compressed MDPs. The FRS tech-
nique estimates an effective route selection metric for
routing the compressed MDPs from PHC station to
CC center.
• Finally, we show that MDC and FRS techniques can be
easily integrated with the existing routing protocols for
WANETs to form an effective RMM system. We use
the simulation to illustrate that the existing protocols for
WANETs with MDC and FRS techniques consume less
energy and reduces packet-loss during the transmission
of the compressed medical data from the PHC station to
CC center.
The rest of the paper is organized as follows: Section II and
Section III present MDC and FRS techniques, respectively.
We present the results of simulation conducted to evaluate the
performance of the work in Section IV and conclude the paper
in Section V.
II. MEDICAL DATA COMPRESSION TECHNIQUE
A WANET is characterized by the frequent occurrence
of transmission errors along with limited battery lifetime.
Another aspect which often concerns is the lossless trans-
mission of MDPs without any perceptual distortions intro-
duced by a compression process and noisy channel. Any
perceptual distortions in MDPs may lead to mis-diagnosis.
Lossless compression techniques are mainly used for medical
image transmission. Such techniques have a constraint of
having a higher bit rate. The visually lossless compression
techniques are therefore widely accepted in medical data
transmission applications. The goal of the MDC technique
is to achieve a highly compressed MDPs without degrading
the perceptual quality and robust against transmission errors.
In this paper, we consider the MDPs in the form of color
images like endoscopic images. The MDC technique is based
on an optimization approach. The block diagram of the
overall compression technique at PHC Station is presented
in Fig. 2. The compression technique is divided into two steps:
image transformation and encoding of coefficients, which
are described in Section II-A and Section II-B, respectively.
To reduce artifacts and noises, the decompressed MDPs are
filtered using an adaptive edge-based fuzzy filter, which is
presented in Section II-D.
www.redpel.com+917620593389
www.redpel.com+917620593389
3. 780 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
A. Image Transformation
The endoscopic images of patients are usually raw color
images [21]. Each pixel in the image is represented by red (R),
green (G), and blue (B) color planes. Independent compres-
sion of the color planes is not efficient as the correlation among
these planes are very high. An adaptive color-space conversion
can perform a better compression. The process of endoscopic
image transformation and quantization is performed in several
processing steps as follows:
1) Color-Space Conversion: The MDC technique performs
an adaptive color-space conversion for the image. Since a
lossless and hardware efficient implementation is required,
YUVr or YCgCo-R yields the best decorrelation of the
image by means of low entropy and without data loss [8],
where YUVr and YCgCo-R correspond to the reversible color
transform of JPEG2000 and the fidelity range extension of the
H.264 video coding standard, respectively. We perform the
color-space conversion from RGB to YCbCr by combining
the aforesaid methods [22] and Structure Conversion [23] for
array squeezing of luminance component.
2) Block-Wise Pixel Scanning: The detailed characteristics
of the image will be omitted after quantization due to the high
computational burden for implementing full-image transfor-
mation. The image is therefore divided into non-overlapping
blocks to decrease the number of operations. The conversion
of progressive pixel scan to block-wise order is required to
operate on small non-overlapping image blocks. The advan-
tages of small block size are less computational complexity
and moderate memory requirement, but the size of compressed
data is large. On the other hand, large block size can attain
significantly compressed data for the consequent coefficients
of small magnitude but may cause visual distortions. However,
to avoid visual artifacts in the case of higher compression of
MDPs, the size of the transformed block was set to 4 × 4. The
MDC technique therefore operates on non-overlapping 4 × 4
image blocks for block-wise data access [10].
3) Transformation and Quantization: The signal dependent
Karhunen-Love transform (KLT) is the most efficient decor-
relating transform for compressing the component planes to
a single spectral plane of a color image prior to encoding
using the maximum energy compaction. However, the KLT
transform is not used in practice because it depends on signal
statistics and does not have an efficient implementation [24].
DCT comes close to ideal transform KLT and more practical
than others [10]. Therefore, 2-D DCT can be used in image
compression process to compact the energy into a few coeffi-
cients along the spatial directions. However, DCT coefficients
are real number and processing these real numbers increases
the computational complexity. We therefore use 2-D integer
DCT (IntDCT) instead of 2-D DCT for transformation and
quantization of the component planes. The degradation in
perceptual quality for using IntDCT is mostly imperceptible
to the human visual system. The MDC technique is thus
considered as visually lossless. Specially, IntDCT provides
good decorrelation property and instruction throughput of
IntDCT can be increased by performing multiple operations
in parallel. Multiplication in the transform process is avoided
by integrating it with the quantization. Block wise access
decreases rounding operation and decreases the computational
overhead. Typical DCT devices are used to implement Int-
DCT. Therefore, redesigning of the hardware is also not
required.
The coefficients obtained from the block transformation
will be entropy encoded using an efficient encoder, which is
presented in the next section.
B. Encoding of Coefficients
In this section, we present a hardware efficient encoder
to encode the transformed coefficients. The coefficients of a
block are partitioned into DC and AC coefficients. First, DC
coefficients are differentially encoded, i.e., the DC coefficient
of the previous block is used as the predicted value for the
current coefficient. DC coefficients of the image can be treated
as a smaller image that essentially has the same smoothness
properties encountered in the image. Next, the differentially
encoded DC coefficients are further encoded using Adaptive
Golomb Rice (AGR) code that uses adaptive coding and
requires only one pass through the data. AGR encoder with a
parameter k is defined by the encoding rule as follows:
G(u, k) = 11 · · ·1
prefix,pbits
0bk−1bk−2 · · · b0
suffixz,kbits
(1)
The AGR encoder encodes the positive integer u as two
strings: a prefix of p + 1 bits, where p = u/2k and a
suffix of k bits. For example, if u = 10 and k = 2, then
the code for u is ‘11010’, the prefix is ‘11’ and the suffix
is ‘10’ [10]. As the AGR encoder encodes only non-negative
integers, the following mapping function is used to transform
DC coefficients to non-negative integers, which is given by
M1(d) =
2d if(d ≥ 0)
2|d| − 1 otherwise,
(2)
where d and |d| denote a DC coefficient and its absolute value,
respectively. AGR decoder will be able to decode correctly the
first block of consecutive error-free codewords in forward and
reverse directions.
Next, the majority of high frequency AC coefficients are
quantized to zero because of the energy compaction property
of IntDCT. The remaining nonzero coefficients in a block
are typically low frequency coefficients clustered around the
DC coefficient. The AC coefficients in are scanned along a
zigzag order. The encoder proposed to encode AC coefficients
uses the principle of the zero run-length encoding. Since AC
coefficients are signed, a mapping function, denoted by M2(e),
is used to transform a nonzero AC coefficient e to a positive
integer for the AGR coding, such that
M2(e) = {M1(e) : e ∈ Z=0}. (3)
A GR code with adjustable k is inefficient when coding low
entropy distributions, such as, encoding zero AC coefficients.
In AGR code, the encoded sequence is therefore rearrange to
form (r, q) pairs, where a nonzero AC coefficient r is followed
by q numbers of consecutive zero-value AC coefficients.
Any zero-value AC coefficient in a block is represented by
a single symbol (0, 0). Since the symbol (0, 0) occurs very
www.redpel.com+917620593389
www.redpel.com+917620593389
4. DUTTA: MEDICAL DATA COMPRESSION AND TRANSMISSION IN WANETs 781
Fig. 3. Block diagram of the adaptive edge-based fuzzy filtering method.
Algorithm 1 Medical Data Compression Technique
1: The color-space conversion from RGB to YCbCr of
an input image is performed to generate less correlated
component planes.
2: Each component plane is transformed and quantized
to reduce inter pixel correlation and packing pixel energy
into few transform coefficients.
2.1: The plane is divided into non-overlapping N × N
blocks.
2.2: The blocks are transformed and quantized by blocked
IntDCT.
3: The quantized coefficients of each component plane are
encoded using a hardware efficient encoder.
3.1: The coefficients of a block are partitioned into DC
and AC coefficients.
3.2: The DC coefficients are differentially encoded and
mapped to non-negative integers using Eq. 2.
3.3: The resultant coefficients are further encoded using
AGR coding shown in Eq. 1.
3.4: The AC coefficients are scanned along a zigzag
order and the nonzero AC coefficients are also mapped
to nonzero positive integers using Eq. 3.
3.5: A pair consisting of a nonzero AC coefficient and the
run-length of the succeeding zero-value AC coefficients is
encoded using the AGR encoder.
3.6: Each zero-value AC coefficient is assigned the
value ‘0’.
frequently, the value ‘0’ is assigned to the symbol (0, 0) for
maximizing the compression efficiency.
C. Medical Data Compression Algorithm
The formal algorithm of MDC technique at PHC Station is
facilitated by Algorithm 1.
In summary, the output of the MDC technique is compressed
MDPs. These compressed MDPs is transmitted from PHC
station of the disaster area to CC center over WANETs
using the FRS technique as described in Section III. At CC
center, the compressed MDPs are decompressed using AGR
decoder followed by the inverse transform and finally inverse
color-space conversion is performed. The AGR decoding and
the inverse color-space conversion are just the reverse process
of AGR encoding (Eq. 1). Due to block transformation,
quantization, and transmission errors due to transmission of
compressed MDPs over noisy channels, blocking artifacts and
different additive and multiplicative noises may add.
Therefore, we introduce an adaptive fuzzy filter during
decompression to reduce such artifacts and noises, which is
described in the next section.
D. Reduction of Artifacts and Noises
The presence of any artifacts in medical images may lead to
mis-diagnosis. Blocking effects result in discontinuities across
block boundaries. In this section, we therefore propose an
adaptive edge-based fuzzy filtering method to reduce blocking
artifacts based on the intensity gradient (slope) of the pixels
close to the boundary of two blocks during decompression
at CC Center. The blocking artifacts are more in lossy com-
pression techniques due to quantization. If the coefficients of
the adjacent blocks are coarsely quantized, a difference in the
intensity gradient across the block boundary is expected. The
strongest filtering that can be applied in medical images are
in the direction perpendicular to the edge. The block diagram
of the adaptive fuzzy filtering method is shown in Fig. 3.
The gradient G of an image I (of size M × N pixels)
is obtained by applying the Sobel kernels, which is a
computational less expensive edge detector, in horizontal
and vertical directions. A global edge map Mg and a
local edge map Ml are obtained by using the edge detec-
tor, such that Mg = 2 GA/M × N and Ml =
(1 − ρ[m, n]/τ[m, n]) × Mg, where ρ[m, n] and τ[m, n]
are the standard deviation and the mean of a block of size
m × n in the gradient image and GA is the amplitude
of the gradient. To be adaptive for different areas having
different activity levels, ρ[m, n] in a window centered on
I(m, n) the spread parameter of the input is defined in [25] as
σ(x[m+m , n+n ], x[m, n]) = S[m+m , n+n ] × σA[m, n],
where value of the amplitude of spread parameter (σA[m, n])
is σ0((1−γ )(ρ(I[m, n]) − ρmin/ρmax − ρmin)+γ ), σ0 is the
maximum spread parameter value, and S is the scaling func-
tion controlled by the direction of x[m+m , n+n ] to x[m, n].
The pixels are first classified into edge pixels and non-
edge pixels by comparing the amplitude of the intensity
gradient to a threshold, denoted by th. Edge pixels are not
further filtered because any smoothing may blur the edges.
For non-edge pixels, if there are no edge pixel in the same
block, a 2-D adaptive low-pass filtering is performed using a
2 × 2 filter over the complete image to remove additive and
multiplicative noises. For non-edge pixels with staircase noise,
the tangent angle of their nearest edge pixel is used to control
the directional spread parameter σ(x[m+m , n+n ], x[m, n]).
For remaining non-edge pixels, the ringing artifacts are filtered
with an isotropic fuzzy filter because such artifacts are not
considered to be oriented in any particular direction [26].
The resultant MDPs obtained at CC center are visually
lossless and ready for diagnosis.
www.redpel.com+917620593389
www.redpel.com+917620593389
5. 782 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
Fig. 4. Block diagram of the FRS Technique.
III. FUZZY-LOGIC BASED ROUTE SELECTION TECHNIQUE
The existing work in the literature on routing in WANETs
does not have any special treatment for the compressed MDPs.
The path generated by the existing routing protocols can
deviate far from the QoS requirements of the compressed
MDPs. We propose FRS technique that reduces the energy
consumption of the network. The FRS technique works with
distance vector routing protocols such as AODV, but it can be
worked with any underlying routing protocol for WANETs.
Our goal is to find a better route selection metric which will
enhance the lifetime of the network.
In this section, we first state the assumptions made about
the network, define the terms and the notations used in this
work, and introduce an energy consumption model. Next, we
propose FRS technique and its application in RMM system.
A. Assumptions
We assume that a large number of nodes, say n, are deployed
in a two-dimensional field of interest (FoI), uniformly at ran-
dom independent of each other. This is a common assumption
in the existing literature, both for theoretical analysis [27] and
on real applications [28]. The uniform random deployment
is favored in a situation where the geographical region to be
routed is hostile and inimical [29]. In an ad-hoc network, if
some nodes go down or in case of node failure, it is also
assumed that the nodes can still send data to the CC center.
We assume the binary disc communication model in which a
node s, can communicate perfectly with other nodes within the
disc of radius Rc centered at s, where Rc is the communication
range of s. The area of the communication region denoted by
A(s, Rc), is nothing but the area of a disk of radius Rc center
at s, i.e., A(s, Rc) = π R2
c . The network is modeled as a
directed graph G = (V, E), where V, |V | = n, is the set of
nodes in the network and E is the edge set. There is a directed
edge (u, v) ∈ E between node u and node v iff single-hop
transmissions from u to v and v to u are possible.
B. Energy Consumption Model
We employ the energy consumption model given in [30].
When a node transmits k bit message directly to a receiver
node over a distance d, the energy consumed for transmission
(Et(k, d)) and reception (Er (k)) can be calculated as follows:
Er(k) = k × Eelect and Et(k, d) = k × (Eelect + εampd2
),
(4)
where the radio dissipates Eelect = 50J/bit to run the
transceiver and receiver circuit and εamp = 100 pJ/bit/m2 for
the transmitter amplifier.
TABLE I
FUZZY RULE SET. I1, I2, I3, LV, VH, H, M, L, AND VL ARE ENERGY
CONSUMPTION, RESIDUAL ENERGY, ROUTING DELAY, LINGUISTIC
VARIABLE, VERY HIGH, HIGH, MEDIUM, LOW, AND
VERY LOW, RESPECTIVELY
C. Overview of FRS Technique
AODV protocol uses hop count as a route selection metric
to find an optimal path in WANETs. However, only the
value of hop count as a route selection metric does not
fulfill the requirements of WANETs. The FL system allows
to combine and evaluate diverse issues of WANETs such as
energy consumption, residual energy, and routing delay in an
efficient manner. The use of FL system is therefore a promising
technique to optimize the route selection metric. The FL sys-
tem performs its work with the help of following four steps:
fuzzifier, fuzzy inference engine, rule base, and defuzzifier.
When an input is applied to a FL system, the fuzzy inference
engine computes the output set corresponding to each rule
base and various methods for inferring the rules. All rules in
the rule base are processed in a parallel manner by the fuzzy
inference engine. The defuzzifier performs defuzzification to
find a single crisp output value. The proposed work considers
m-input 1-output FL system using singleton fuzzifica-
tion, center-of-sets defuzzification, and IF-THEN rule. The
FL system uses the following IF-THEN rule: IF input1
→ x1 and input2 → x2 and · · · inputm → xm, THEN
output → y.
The FRS technique uses a FL system to compute the
route selection metric as shown in Fig. 4. The inputs are
energy consumption, residual energy, and routing delay and
the output, denoted by select_routing_metric, is the probability
of a path to be selected for routing.
A path having higher value of select_routing_metric will
have more chances to be a part of the routing path. The
FL system expresses numeric data in word language called
linguistic variable (LV). The values of LVs are words or
sentences in a natural or artificial language providing a means
of systematic manipulation of vague and imprecise concepts.
Table I illustrates the LVs for inputs of FL system.
Inputs of the FL system are described as follows:
1) Energy Consumption: An energy-efficient path in route
selection phase can prolong the lifetime of WANETs. Let Eab
ik
www.redpel.com+917620593389
www.redpel.com+917620593389
6. DUTTA: MEDICAL DATA COMPRESSION AND TRANSMISSION IN WANETs 783
Fig. 5. An example of FRS technique. Label on an edge ad and value at
node a are Ead
il and the residual energy of a, respectively.
denotes energy required to transmit k bit message of com-
pressed MDPs between nodes a and b via ith path, where
i ≤ t and t is the number of possible paths between a and b.
From Eq. 4, Eab
ik is therefore given by
Eab
ik =
hi
j=1
k × (Eelect + εamp hij
2
) + k × Eelect , (5)
where hi and hij are the hop count and length of the jth
hop in the ith path, respectively. Fig. 5 illustrates a scenario
where s and t are source and destination nodes, respectively.
A relay node c receives two RREQ messages from nodes b
and d. By using Eq. 5, the actual energy required to transmit
data between s and c can be written as Esc
1k = {2 + 2 + 2}
and Esc
2k = {2 + 3}. The energy required in an energy-
efficient path between s and c, denoted by Esc
k , is expressed as
Esc
k = min{Esc
1k, Esc
2k, . . . , Esc
tk }. The LVs for energy consump-
tion are high, medium, and low. The values of LVs are nothing
but the energy required to transmit the compressed MDPs
between nodes a and b via ith path, i.e., Eab
ik .
2) Residual Energy: Considering the residual energy of
nodes in route selection metric can avoid nodes from being
overused and eventually lead to an increase in the operational
lifetime of WANETs. Let eab
ij denotes the energy of the jth
node in ith path between nodes a and b, where i ≤ t,
j ≤ hi , t is total paths between a and b, and hi is the
hop count in ith path. The residual energy in the ith path is
given by
eab
i = min{eab
i1 , eab
i2 , . . . , eab
ihi +1}. (6)
Let us continue the previous example, as illustrated in
Fig. 5 to emphasize the residual energy fuzzy set. By using
Eq. 6, the residual energy between nodes s and c are esc
1 =
min{5, 6} and esc
2 = {2}. The LVs for residual energy are high,
medium, and low, where high ≥ Einit − Eth, medium ≥
(Einit ± Eth)/2, low ≥ Eth, Einit is the initial energy of
nodes, and Eth is the minimum residual energy required to
transmit compressed MDPs. Let k is the size of the compressed
MDPs. By using Eq. 4, the value of Eth can be written as
Eth = k × (Eelect + εamp × C2) + k × Eelect .
3) Routing Delay: WANETs in case of an emergency situa-
tion will not be delayed beyond a predefined delay threshold.
Let dab
i denotes the routing delay to transmit data between
nodes a and b via ith path, where i ≤ t and t is the number of
possible paths between a and b. The value of dab
i depends on
propagation delay, transmission delay, and processing delay
denoted by dpn, dtr, and dpg, respectively. We assume that
dpn +dtr +dpg= f , where f is a constant value for equal length
hop in the routing path. Let hi be the hop count between nodes
Algorithm 2 Fuzzy-Logic Based Route Selection
Technique
1: When a node d wants to find a route to destination
node t, it broadcasts RREQ messages {s,t,d, Esd
k , esd
1 , dsd
1 }.
2: If an intermediate node c receives RREQ messages for t:
2.1: Node c uses the fuzzy rule set of Table I and calculates
the select_routing_metric for each RREQ message.
2.2: Node c selects RREQ message having least
select_routing_metric.
2.3: Node c updates its routing table and broadcasts a
RREQ message.
3: When t receives RREQ messages:
3.1: Node t uses fuzzy rule set (Table I), calcu-
lates select_routing_metric for each RREQ message, and
updates its routing table.
3.2: t waits for a fixed time interval to receive more route
request messages.
3.3: Node t unicasts a RREP message back to
its neighbor from which it has received the least
select_routing_metric.
4: Each node, after receiving a RREP message, unicasts
the RREP message towards source node s.
a and b via ith path. The routing delay can be expressed as
dab
i = hi ×(dpn +dtr +dpg) = hi × f . Fig. 5 illustrates that the
values of dsc
1 and dsc
2 are 3 f and 2 f , respectively. The LVs
for routing delay are high, medium, and low. The values of
LVs are nothing but the delay to transmit compressed MDPs
between nodes a and b via ith path.
Next, the value of select_routing_metric is calculated using
the following if-then rule: IF energy consumption → Low
and residual energy → High and delay → Low, THEN
the select_routing_metric → Highest. The fuzzy mapping
rules are obtained based on three fuzzy inputs and their
corresponding LV as illustrated in Table I.
D. Application of the Proposed FRS Technique
The proposed FRS technique can be integrated with any
underlying routing protocol. In this section, we describe the
AODV_FLT algorithm that uses select_routing_metric as route
selection metric to find an effective routing path for WANETs.
An example of the AODV_FLT algorithm is shown in Fig. 5.
The network consists of seven backbone nodes {s, a, b, c, d,
e, t}, where s and t are the source and the destination, respec-
tively. The label on an edge
−→
ab shows the energy consumption
to transmit k bit message of compressed MDPs between nodes
a and b. An entry in the routing table at a node d shows the
route information i.e., [s,t,a, Esa
k + Et(k, ad ), min{esa
1 , ea},
dsa
1 +1], where a and ea are the predecessor neighbor of d
and residual energy of a, respectively. Fig 5 illustrates that
the routing table at a node d is therefore {s, t, a, 4, 5, 2}. The
RREQ message broadcasts from d is denoted by {s,t,d, Esd
k ,
esd
1 , dsd
1 }. Fig 5 illustrates that the RREQ message broadcasts
from d is {s, t, d, 4, 5, 2}. A relay node c receives RREQ
messages from nodes d and b. Node c uses Table I to select
an optimal route request.
www.redpel.com+917620593389
www.redpel.com+917620593389
7. 784 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
Fig. 6. Original test endoscopy images (first row) of the dataset [33], respective histograms (second row), their decompressed counterpart (third row) and
their respective decompressed histograms (last row).
E. Fuzzy-Logic Based Route Selection Algorithm
The formal algorithm of AODV_FRS technique is facilitated
by Algorithm 2.
F. Complexity Analysis of Algorithm 2
1) Message Complexity: Each node broadcasts a RREQ
message to its neighbors and sends a RREP message to
its predecessor neighbor node. Each node receives RREQ
messages from its neighbors and one RREP message from
its successor node. Let x is the number of neighbors
of a node. Since x N, the message complexity of
Algorithm 2 is O(N × (x + 1 + x + 1)) = O(N) which is
nearly optimal.
2) Space Complexity: We assume a routing table consists
up to three best possible path information. This is an assump-
tion widely used in the literature [11]. A routing table at
node d has the following fields: s, t, b, Esb
k + Et(k, bd ),
min{esb
1 , eb}, and dsb
1 + 1. The space complexity is therefore
O(N × 3 × 6) = O(N) which is nearly optimal.
IV. SIMULATION RESULTS
In this section, we discuss the results from a simulation
study of the proposed MDC and FRS techniques imple-
mented in MATLAB 7.1 [31] and NS 2.34 simulation [32],
respectively. The MDC technique generates compressed MDPs
at PHC station from raw endoscopic images of patients. The
compressed MDPs are decompressed and artifacts and noise
are removed before diagnosis at CC Center. These compressed
MDPs are then transmitted to CC Center via WANETs. The
PHC station and CC Center are therefore working as source
and destination nodes in the FRS technique.
A. Performance Evaluation of the MDC
This section evaluates the performance of the proposed
MDC technique and compares to JPEG, JPEG2000, and [10]
compression schemes. An image dataset of 500 endoscopic
images with a 512 × 512 resolution [33] are used for compres-
sion and are tested against transmission errors, such as least
significant bit error. Unlike [9], [10], the MDC technique can
be effectively used for compression of other forms of medical
images, e.g. electrocardiography images, magnetic resonance
imaging, etc.
1) Visual Quality of MDPs: Fig. 6 illustrates that the
degradation in visual quality of the compressed MDPs is
not perceived by the human visual system. In addition, the
histograms of original and decompressed images are same with
high probability. Peak Signal-to-Noise Ratio (PSNR) is used
to evaluate the quality of image reconstruction, i.e., PSNR
(d B) = 10log(2552/(ai − ¯ai)2), where ai and ¯ai are values
of a pixel in original and decompressed images, respectively
and · is the averaging operator. Mean Opinion Score (MOS) is
the average quality rating over a number of human observers
that have been asked to score an image, often on the scale
from 1 (worst) to 5 (best) [34]. An objective MOS prediction
uses perceptual metrics that correlate with human perception
of image quality using blockiness and blur in combination with
a few others to estimate the perceived quality of an image.
2) Compression Ratio: Compression Ratio (CR) is the ratio
of the size of an original image to the size of its compressed
counterpart. As the CR increases, the bit rate will decrease
to achieve highly compressed data. However, visual artifacts
may occur when CR is very high in case of lossy compression
techniques. Fig. 7(a) illustrates the efficiency of Algorithm 1
in terms of MOS for endoscopic images of the dataset [33] in
the noiseless transmission channel. DCT based compression
methods, such as JPEG, MDC, and [10], provide better
perceptual quality when CR is less compared to DWT
based methods, like JPEG2000. Blockiness effect increases
in DCT based method with the increase in CR which in turn
degrade the perceptual quality of the compressed images [34].
No such blocking artifacts occurs in DWT based meth-
ods. Therefore, JPEG2000 provides better perceptual qual-
ity then JPEG and [10] at higher CR. Furthermore, AGR
www.redpel.com+917620593389
www.redpel.com+917620593389
8. DUTTA: MEDICAL DATA COMPRESSION AND TRANSMISSION IN WANETs 785
Fig. 7. Illustration of Perceptual quality of compressed MDPs. (a) Perceptual
quality of MDPs degrades with compression ratio. (b) Perceptual quality
od MDPs in a noisy channel. (c) Visual quality of MDPs in presence of
transmission errors.
(MDC and [10]) encoder provides better perceptual quality
with higher compression than Huffman (JPEG) and Bit plane
encoder (JPEG2000) as illustrated in literature [10]. The
proposed adaptive edge-based filtering in MDC takes care
of blocking artifacts and quantization noise. Therefore, the
proposed MDC technique is less sensitive to changes in CR
in the noiseless channel.
3) Robustness Against Transmission Errors: The bit error
rate (BER) is the ratio of the number of error bits to the
total number of transferred bits during a studied time inter-
val. BER against transmission errors can be calculated as
BER = transmission errors/ total number of bits. Robust-
ness are estimated based on BER, such that robustness =
(1-BER)×100%. Figs. 7(b) and 7(c) justify that Algorithm 1
maintains higher MOS and PSNR, respectively, even when
the BER is high. MOS and PSNR of the MDPs drop with
the increase in BER due to increase in noise addition, which
may introduce artifacts. DWT based method (JPEG2000)
provides better visual quality than DCT based techniques
(JPEG and [10]) when the BER is high. In [10], when the BER
is greater than 0.15, the PSNR drops drastically degrading
the visual quality of MDPs. However, JPEG2000 is also
sensitive to transmission errors. The perceptual quality of the
JPEG2000 encoded MDPs degrades when the BER is greater
than 0.1. An adaptive edge-based fuzzy filter is described
in Section II-D, which relatively reduce the effect of the distor-
tion on the block borders, will protect the detailed edges from
blurring, and minimizes the impact of other multiplicative
and additive noise, such as transmission errors. In short, the
impact of increasing BER is less in case of the MDC, which
eventually provides better perceptual quality even when the
MDPs are transmitted through noisy channels.
Fig. 8. Illustration of the lifetime of network. (a) Relationship between
network lifetime and remaining energy. (b) Relationship between number of
sensors and network lifetime. (c) Relationship between network lifetime and
delivery ratio of MDPs.
B. Performance Evaluation of the FRS
We deployed 400 sensors uniformly at random independent
of each other in a square-shaped FoI of 1000m × 1000m.
The destination is located at the position (990, 990). The
initial energy and the communication range of the sensors
are assumed to be 25J and 100m, respectively. We have
implemented FRS technique on AODV routing protocol with
and without using MDC technique. The enhanced versions of
AODV are referred to as AODV_FRS, AODV_FRS_MDC,
and AODV_MDC. The goal of this simulation is to show
how the FRS technique prolongs the network lifetime without
compromising the QoS of WANETs.
1) Energy Consumption of Network: First, we study the
impact of the proposed work on the energy consumption of
the network. We measure the total energy consumption of
network with the simulation time. Fig. 8(a) shows the total
energy consumption of the network for the entire duration of
the simulation. It can be concluded from the result in Fig. 8(a)
that when proposed MDC technique is not used, all the
uncompressed MDPs are routed to the destination and the
energy consumption is then very high. Fig. 8(a) also shows that
AODV_FRS consumes less energy than simple AODV. This is
because AODV routing without the proposed FRS technique
requires more number of hop count for routing MDPs.
2) Lifetime of Network: Next, we study the impact of the
proposed technique on the lifetime of the network. We measure
the network lifetime with the number of rounds of simulation
time till all the nodes drain their energy completely. Fig. 8(b)
shows the relationship between the network lifetime and
the number of nodes. It can be observed from the results
in Fig. 8(b) that when the MDC technique is not used,
www.redpel.com+917620593389
www.redpel.com+917620593389
9. 786 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
uncompressed MDPs are routed to the destination and then
the energy of the nodes drain out rapidly. It also shows that
when the proposed FRS technique is used, the energy of the
nodes is equally consumed, because residual energy becomes
a part of route selection metric.
3) Delivery Ratio of MDPs: Finally, we show the impact of
the proposed work on MDPs delivery ratio. The MDPs deliv-
ery ratio of a flow is the ratio of the number of MDPs that are
received by the destination over the number of MDPs sent
by the source. Fig. 8(c) shows the overall MDPs delivery
ratio for the entire duration of the simulation. AODV_FRS,
AODV_FRS_MDC, and AODV_MDC achieve a stable perfor-
mance at the entire duration of the simulation, because route
length becomes a part of route selection metric. It illustrates
that the delivery rate is inversely proportional to the route
length (hop count).
V. CONCLUSION
In this paper, we proposed the RMM system for routing
the medical data of patients in the disaster area. The proposed
system comprises a set of components which collects, com-
presses, and transmits compressed medical data to the base
station using WANETs. The coding technique in RMM system
allows to decode correctly even in the presence of transmission
errors. RMM system exploits the attributes of WANETs to
maintain QoS of WANETs. We simulated the performance
of the RMM system for different network scenarios and
demonstrated that the lifetime of the network is increased
due to the routing of compressed MDPs and hence maintains
QoS of WANETs. In our future research, we intend to make
hardware implementations. We believe that this work also
motivates further research in the security problem in WANETs.
REFERENCES
[1] Uttarakhand Flood. [Online]. Available: http://en.wikipedia.org/wiki
/2013_North_India_floods, accessed 2014.
[2] B. Fong, N. Ansari, and A. C. M. Fong, “Prognostics and health man-
agement for wireless telemedicine networks,” IEEE Wireless Commun.,
vol. 19, no. 5, pp. 83–89, Oct. 2012.
[3] K. Kang, J. Ryu, J. Hur, and L. Sha, “Design and QoS of a wireless
system for real-time remote electrocardiography,” IEEE J. Biomed.
Health Inform., vol. 17, no. 3, pp. 745–755, May 2013.
[4] S. M. George et al., “DistressNet: A wireless ad hoc and sensor
network architecture for situation management in disaster response,”
IEEE Commun. Mag., vol. 48, no. 3, pp. 128–136, Mar. 2010.
[5] Y. Gadallah and M. A. Serhani, “A WSN-driven service discovery
technique for disaster recovery using mobile ad hoc networks,” in Proc.
IFIP Wireless Days, Oct. 2011, pp. 1–5.
[6] K. Shimazu, M. Ito, Y. Kitsuta, and N. Kawashima, “Project concept:
Design concept of ad-hoc information network system for disaster
mitigation,” in Proc. ICACCI, Aug. 2013, pp. 1436–1439.
[7] L. Guo, S. Umbaugh, and Y. Cheng, “Compression of color skin tumor
images with vector quantization,” IEEE Mag. Eng. Med. Biol., vol. 20,
no. 6, pp. 152–164, Nov./Dec. 2001.
[8] G. Schaefer and R. Starosolski, “An evaluation of image compression
algorithms for colour retinal images,” Biological Med. Data Anal.,
vol. 4345, 2006, pp. 358–367.
[9] C. Cheng, Z. Liu, C. Hu, and M. Q.-H. Meng, “A novel wireless
capsule endoscope with JPEG compression engine,” in Proc. IEEE ICAL,
Aug. 2010, pp. 553–558.
[10] P. Turcza and M. Duplaga, “Hardware-efficient low-power image
processing system for wireless capsule endoscopy,” IEEE J. Biomed.
Health Inform., vol. 17, no. 6, pp. 1046–1056, Nov. 2013.
[11] C. E. Perkins and E. M. Royer, “Ad-hoc on-demand distance vector
routing,” in Proc. IEEE WMCSA, Feb. 1999, pp. 90–100.
[12] S. Ziane and A. Mellouk, “Inductive routing based on dynamic end-to-
end delay for mobile networks,” in Proc. IEEE GLOBECOM, Dec. 2010,
pp. 1–5.
[13] W. El-Hajj, A. Al-Fuqaha, M. Guizani, and H.-H. Chen, “On efficient
network planning and routing in large-scale MANETs,” IEEE Trans.
Veh. Technol., vol. 58, no. 7, pp. 3796–3801, Sep. 2009.
[14] K. Namuduri and R. Pendse, “Analytical estimation of path dura-
tion in mobile ad hoc networks,” IEEE Sensors J., vol. 12, no. 6,
pp. 1828–1835, Jun. 2012.
[15] G. Varaprasad, “High stable power aware multicast algorithm for mobile
ad hoc networks,” IEEE Sensors J., vol. 13, no. 5, pp. 1442–1446,
May 2013.
[16] J. Vazifehdan, R. V. Prasad, and I. Niemegeers, “Energy-efficient reliable
routing considering residual energy in wireless ad hoc networks,” IEEE
Trans. Mobile Comput., vol. 13, no. 2, pp. 434–447, Feb. 2014.
[17] JPEG. [Online]. Available: http://http://www.jpeg.org/, accessed 2014.
[18] JPEG2000. [Online]. Available: http://www.jpeg.org/jpeg2000/,
accessed 2014.
[19] O. Deforges, M. Babel, L. Bedat, and J. Ronsin, “Color LAR codec:
A color image representation and compression scheme based on local
resolution adjustment and self-extracting region representation,” IEEE
Trans. Circuits Syst. Video Technol., vol. 17, no. 8, pp. 974–987,
Aug. 2007.
[20] J.-J. Ding, Y.-W. Huang, P.-Y. Lin, S.-C. Pei, H.-H. Chen, and
Y.-H. Wang, “Two-dimensional orthogonal DCT expansion in trapezoid
and triangular blocks and modified JPEG image compression,” IEEE
Trans. Image Process., vol. 22, no. 9, pp. 3664–3675, Sep. 2013.
[21] R. Melo, J. P. Barreto, and G. Falcao, “A new solution for camera cali-
bration and real-time image distortion correction in medical endoscopy–
initial technical evaluation,” IEEE Trans. Biomed. Eng., vol. 59, no. 3,
pp. 634–644, Mar. 2012.
[22] S.-Y. Lee and A. Ortega, “A novel approach of image compression in
digital cameras with a Bayer color filter array,” in Proc. ICIP, vol. 3.
2001, pp. 482–485.
[23] C. C. Koh, J. Mukherjee, and S. K. Mitra, “New efficient methods of
image compression in digital cameras with color filter array,” IEEE
Trans. Consum. Electron., vol. 49, no. 4, pp. 1448–1456, Nov. 2003.
[24] U. Bayazit, “Adaptive spectral transform for wavelet-based color image
compression,” IEEE Trans. Circuits Syst. Video Technol., vol. 21, no. 7,
pp. 983–992, Jul. 2011.
[25] D. T. Võ, T. Q. Nguyen, S. Yea, and A. Vetro, “Adaptive fuzzy filtering
for artifact reduction in compressed images and videos,” IEEE Trans.
Image Process., vol. 18, no. 6, pp. 1166–1178, Jun. 2009.
[26] A. S. Al-Fohoum and A. M. Reza, “Combined edge crispiness and
statistical differencing for deblocking JPEG compressed images,” IEEE
Trans. Image Process., vol. 10, no. 9, pp. 1288–1298, Sep. 2001.
[27] K.-S. Hung and K.-S. Lui, “On perimeter coverage in wireless sensor
networks,” IEEE Trans. Wireless Commun., vol. 9, no. 7, pp. 2156–2164,
Jul. 2010.
[28] L. Mo et al., “Canopy closure estimates with GreenOrbs: Sustainable
sensing in the forest,” in Proc. ACM Conf. ENSS, 2009, pp. 99–112.
[29] A. Akhter and T. Sanguankotchakorn, “Modified AODV for multi-
constrained QoS routing and performance optimization in MANET,”
in Proc. ECTI-CON, May 2010, pp. 234–238.
[30] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-
efficient communication protocol for wireless microsensor networks,”
in Proc. Hawaii ICSS, vol. 2. 2000, pp. 1–10.
[31] MATLAB. [Online]. Available: http://www.mathworks.in/,
accessed 2014.
[32] The Network Simulator—NS-2.34. [Online]. Available: http://www.
isi.edu/nsnam/ns, accessed 2014.
[33] Gastrolab. [Online]. Available: http://www.gastrolab.net, accessed 2014.
[34] F. Ebrahimi, M. Chamik, and S. Winkler, “JPEG vs. JPEG2000: An
objective comparison of image encoding quality,” Proc. SPIE ADIP,
vol. 5558, pp. 300–308, Nov. 2004.
Tanima Dutta received the B.Tech. degree from
the Haldia Institute of Technology, India, in 2005,
and the M.Tech. degree from Calcutta University,
Kolkata, India, in 2010. She is currently pursing the
Ph.D. degree in computer science and engineering
from Indian Institute of Technology Guwahati, India.
She is currently with TCS Innovation Labs, Ban-
galore, India. Her research interests include wire-
less networks, multimedia, computer vision, and
algorithms.
www.redpel.com+917620593389
www.redpel.com+917620593389