SEMINAR 2014

ALPR

1. INTRODUCTION

Chronic pain impactsover100million Americans with annual treatment costs exceeding $100
billion [1], [2]. One treatment for severe chronic pain is the implantation of an intrathecal drug delivery
device (IDDD) [3][7]. Conventionally, an IDDD that allows active control of the delivery rate is
composed of several components other than the drug reservoir, including a peristaltic pump, a control
circuit, and a battery.The inclusion of these components reduces the fractional volume of the unit that may
be occupied by the drug itself; this is considered the volume efciency ratio (VER). An alternative
architecture that uses throttle valves to regulate ow from pressurized drug reservoirs can potentially
improve the VER by obviating the pump [8]. This architecture is also appealing for multidrug delivery
systems. Delivery of multiple drugs into the intrathecal space is utilized in clinical practice when
monodrug therapy with opioids fails, either due to tolerance, opioid nonresponsiveness (neuro- pathic
pain), or hyperalgesia [9], [10]. Current practice consists of mixing multiple drugs in a single chamber for
simultaneous deliveryintospina luid[11][13].Thisconstrainsthetypeand concentration of drugs that can
be administered. In this paper, we explore an IDDD system architecture that utilizes a silicon
micromachined two-valve manifold to reg- ulate ow from pressurized reservoirs. Piezoelectric actuation
is used for power efciency. The manifold also in- cludes three pressure sensors that may be used to
determine delivery rates.1 A system with multiple independently regu- lated drug reservoirs can improve
analgesic

efcacy,

reduce

treatmentcost,andultimatelyimprovepatientsatisfaction.This

method

of

combination therapy also limits side effects because it uses the lowest possible doses of each drug.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

2. A Multidrug Delivery System

Fig: 1 System overview

A drug delivery system for chronic pain that can accommodate multidrug protocols. An element that is
important to the function of the system is a customized silicon micromachined valve manifold. Two

polymer reservoirs are pressed by a plate attached to compressive springs to generate pressure.
Two valves in a manifold are used to regulate drug delivery rates. Control is regulated by
onboard electronics that allow for pressure monitoring and reprogramming.

3. MANIFOLD DESIGN AND TESTING

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

The valve manifold should modulate liquid ow from less than 0.5mL/day to more than
5mL/day from each reservoir independently, although typical delivery rates average near the
middle of that range. Both the valve and the embedded pressure sensors should be able to
accommodate pressures as high as 100 kPa. (The precise load will depend on the targeted range
of delivery pressure and the drug volume remaining in the reservoir.) While actuation
mechanisms used in microsystems range from electrostatic to thermal phase change, this system
favors piezoelectric actuation because this approach consumes low power and generates forces
that allow valve operation acrossawiderangeofoperatingpressures[15][24].Asilicon- on-insulator
(SOI) wafer is used for the manifold to ease the fabrication of embedded sensors. The pressure
sensors use implanted piezoresistors and target a sensitivity of at least 600 ppm/kPa for accurate
feedback control. The manifold (Fig. 2) operates by pressing elongated silicon valve seats against
a glass wafer using piezoelectric actuators. The elongated valve seat increases the ow perimeter
which compensates for the limited deection of the piezoelectric actuators [25]. The
piezoelectric actuators used in the manifold are commercially available lead zirconatetitanate
(PZT) stacks. The valves throttle by constricting the ow path from the reservoirs to the
intrathecal space. The 02-m spacing between the valve seat and the glass plate creates a
channel that has a much higher hydraulic resistance than the delivery catheter.The small size and
low power consumption of the manifold allow for larger reservoirs and smaller batteries to be
used without increasing the size of an implantable system.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

Fig. 2. On the top side of the Si wafer, which is the substrate side of an SOI
wafer, there are two membranes with serpentine valve seats. Each one is
located on a central boss that also houses a pressure sensor and presses
against a glass plate that has two inlets and an outlet.The back side, which is
the epitaxial sideof the SOI wafer, has boron doping and gold traces for the
pressure sensors.
.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

3.1 FABRICATION

The manifold is fabricated from an SOI wafer with layer thicknesses of 20, 2, and 450 m and
a 500-m-thick Pyrex glass wafer. The SOI and glass wafers are independently processed and
bonded prior to singulation. The fabrication processes for the SOI and glass wafers are shown in
Fig. 3. For the SOI wafers, the rst step is to form the piezoresistors for the pressure sensors on
the device side of the wafers. Boron implantation is performed with a 11013 cm2 dose at an
energy level of 20 keV with a 7 tilt. The implantation, drive-in, and annealing temperatures
were calculated using the TSUPREM-4 implantation simulator and selected to yield a carrier
prole that provides at least 600-ppm/kPa sensitiv- ity. The simulations indicate that the resistors
should have a sheet resistance of about 5.3103 /and a resistance of 112.6 k.
After the boron implantation, a thermal oxidation step pro- vides both the necessary isolation
and the high-temperature anneal needed to activate and diffuse the implanted boron. A 1000-thick oxide is grown at 1000 C for 20 min, and he wafer is left in a nitrogen environment for an
additional 5 min to allow enough diffusion time for the implanted boron. Vias are etched in the
oxide using buffered hydrouoric acid (BHF) to create contacts to the buried piezoresistors, and
metal is deposited on the wafer to make ohmic contact with the boron- doped silicon.A
platinum/gold(500/1000--thick) layer with a titanium adhesion layer(50)isevaporated on to
the wafer and put in acetone to pattern via lift off. Lastly, more gold (1000 ) is evaporated with
a chrome adhesion layer (300 ) to form the conductive traces and contact pads for easier
electrical interface with external wiring. Fig. 4 shows the fabricated manifold with a standard
starburst pattern.
After the device side is processed, the back side of the wafe rissubjected to two deepreactiveionetch(DRIE) process steps to create the pressure sensor diaphragms, the valve mem- brane
suspension, and the serpentine grooves that provide the elongated ow perimeter. The two-step
DRIE process uses aluminum and photoresist masks to achieve the desired groove and
membrane structures. The resulting grooves are 120 m in depth and provide a seat perimeter
that is 81 mm long. (Other versions of this valve had perimeter lengths of 315 and 14 mm; under
DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

similar operating conditions, these would be appropriate for higher and lower ow rates,
respectively).
A Pyrex glass wafer is patterned with a 1.4-m recess using BHF and is then subjected to a HF
etch to form through vias that act as the valve inlets and outlets. After the recess is formed, both
sides of the glass wafer are patterned with chrome, gold, and a thick photoresist. The wafer is
then placed in HF until the through holes are etched. After stripping the etch mask, a thin lm
layer of sacricial aluminum is deposited through evap- oration into the recessed areas to prevent
inadvertent bonding in the next step. The SOI and glass wafers are then anodically bonded at 400
C, and the wafers are diced to create the nal valve die.
After the valves are fabricated, solder connections are made to the pressure sensor pads,
and they are assembled with the PZT actuator stack inside a ceramic Macor cap. The valve plates
on the manifold can be individually adjusted during the bond step to create a normally open, a
partially open, or a normally closed valve. More information on this method of assembly is
provided in [26]. The nal manifold structure measures 31.51.1 cm3 when assembled with a
Macor header to provide uid interconnects. The assembled structure is shown in Fig. 4.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

Fig. 3. Siglass micromachining process.

Sensors are formed on the device layer of the SOI wafer by implanting
boron and making ohmic connections through a field oxide to gold contact
pads on the device perimeter. The buried oxide layer in the SOI wafer acts as
an etch stop for DRIE when forming membranes. A two-step DRIE process is
illustrated for the SOI wafer. A glass wafer undergoes two wet etch steps for
arecess and through-hole formation.Next, the two wafers are bonded and
diced.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

Fig. 4. Photographs of a fabricated valve manifold.

(a) The glass inlets and outlets, the two valve seats, and the pressure
sensor cavities are
clearly visible onthe top.
(b)The gold contact pads and traces for the pressure sensors and the
gold to define the
PZT locations are clearly visible on the back side of the device.
(c) Both valve plates are clearly visible, and the common outlet port is
located in the
center of the die. The manifold measures 3 1.5 1.1 cm3.
(d)Inlets and outlet tubes of the assembled header interface with the
valve through a
Macor connector piece.

DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

3.2. EXPERIMENTAL RESULTS

Partially open manifold valves were tested for gas and liquid regulation at room
temperature. Gas flow experimentation wasconducted using pressurized nitrogen and is
presented in [27].For liquid modulation testing, experiments were conductedusing regulated
nitrogen gas to pressurize a liquid reservoir.Pressure-driven isopropyl alcohol (IPA) flow was
regulatedinto the individual inlets of the manifold by external valves[Fig. 5(a)]. Flow leaving the
valve manifold (or single valve)was determined by recording the travel of air bubbles alonga 1m-long catheter at the outlet of the device. The catheteris similar in length and size to those used
in intrathecal drugdelivery implants. This mimics the output resistance experiencedin the drug
delivery system and allows for more accuratecharacterization of the liquid flow rate as it pertains
to thetargeted system application.

3.3. LIQUID FLOW TESTS

Liquid modulation tests were conducted on the manifolds to determine flow properties
and

mixing from the independently regulated reservoirs.In a typical test, flow from valve A

(one side of the manifold) varied from 1.77 to 0.028 mL/h, and flow from valve B varied from
2.12 to 0.38 mL/h in an 80-V actuation range [Fig. 5(b)]. When both valves A and B were open
together, the flow rates were combined from each valve within two separate flow regimes. At
high flow rates, the catheter resistance dominated the flow. At lower flow rates, the valve
resistance dominated,and the combined flow rates wereclose to the individual flow rates of each
valve added together.
Several piezoresistors were tested at room temperature, and the resistance varied
between 100 and 150 k, indicating close correspondence with the simulation. The pressure
sensors were driven at 5 V as the differential output voltage was monitored.Multiple pressure
sensors were monitored over 3 d to determine linearity and sensitivity. At room temperature, the
DEPT OF ECE

MZC KADAMMANITTA

SEMINAR 2014

ALPR

pressure sensors showed a sensitivity of 698 ppm/kPa, which exceeds our minimum target for
this application.

Fig.5
(a) Valve test setup. Nitrogen is used to pressure fluid in a large reservoir. The pressurized fluid
is independently routed to each side of the manifold, and a video camera records the travel
of bubbles through a 1-m catheter tube to determine the flow rate through the valve.
(b) Flow rate of alcohol through individual and combined valves in a manifold
at 14 kPa. The
operating region. This results in reduced modulation. Mixing also
functions as expected and
DEPT OF ECE
KADAMMANITTA

10

MZC

SEMINAR 2014

ALPR

is particularly catheter resistance is greater than the valve resistance at

low voltages outside
the preferred evident in the lower flow rate regimes.

4. SYSTEM RESULTS
This section describes the system assembly, flow tests, and safety experiments. A prototype
of a drug delivery system is shown in Fig. 6. The system was operated using a custom designed
control printed circuit board that operated from a3.3-V battery. Flow control from multiple
reservoirs was tested using both continuous regulation and duty-cycle regulation.
Failure mode testing indicates that embedded pressure sensors can be used to detect certain
failures that are possible during deployment.

4.1. HOUSING,ASSEMBLY,AND COMPONENTS

The system requires development of a biologically compatible housing with integrated
components that are necessary for delivery monitoring and regulation. This prototype is made
from stainless steel using traditional machining. An actual realization of the IDDD would likely
be made of biocompatible titanium. The stainless steel housing contains a two-valve manifold, a
control circuitry, two polyethylene (PE) reservoirs, a steel pressure plate, traditional springs, and
a battery. The system housing measures 5.08 cm 9 cm 3 cm, with 5-mm beveled edges. The
housing has a total volume of 130 cm3 with a total reservoir volume of 40 cm3. This prototype
has a VER of 30.7%, but it can be increased by integrating the system components into the
housing more efficiently.
The drug delivery system has three ports. Two of these are standard refill ports (Instech
Laboratories, Plymouth Meeting, PA; model MIC) that are connected to the two reservoirs by
DEPT OF ECE
KADAMMANITTA

11

MZC

SEMINAR 2014

ALPR

tubing. The ports have a silicone elastomer septum, an open cavity, and a metal base plate and
are integrated into the system casing. These ports permit the reservoirs to be filled by a syringe
equipped with a Huber needle. The third port is a fluidic access port. The port is connected
directly to the catheter to allow a physician to circumvent the device for direct access to the
cerebrospinal fluid.
The PE reservoirs in this prototype are pressurized through compression by a metal plate
[Fig. 7(a)]. The reservoirs have an inlet tube from the refill port and exit tubing to the manifold
and its inlet pressure sensors. The reservoirs are fabricated by heat-sealing PE into the desired
shape to fit within the housing cavity designed for the pressure plate. The reservoirs are stacked;
compressed springs drive the pressure plate against the top reservoir, which, in turn, presses
against the bottom reservoir. There are vertical guides in the plate and in the casing to maintain
alignment of the plate, springs, and reservoirs during the assembly and compression. The
pressure response of the reservoir can be adapted to specific system needs by appropriately
selecting the compressive springs used to drive the pressure plate. In this case, two centralized
compressive springs pressurized the reservoirs. The springs were characterized by filling the
reservoirs with a control volume of liquid (1 mL, IPA) and measuring the resultant pressure
using the internal pressure sensors. (For this test, the reservoir outlets were sealed to prevent
outflow.) The pressures varied from 0 kPa when completely empty to 5.17 kPa at 10 mL of total
volume[Fig. 7(b)]. This represents a typical increase of 0.52 kPa/Ml which is acceptable for
system functionality.

DEPT OF ECE
KADAMMANITTA

12

MZC

SEMINAR 2014

ALPR

(a) Photograph of a reservoir before it is inserted into the system. The

reservoir is connected to the refill port and interconnect tubing that
leads to the manifold and a pressure sensor.
(b)Results from one reservoir filling test where each reservoir was filled
with IPA in 1-mL increments. The results for fill volumes up to 10 mL
suggest a linear pressure profile of 0.52 kPa/mL.

4.2. EXPERIMENTAL METHODS

The experimental evaluation method allowed the entire flow path to be tested from the
refill ports, through the system, to a delivery catheter with a pressure load that mimicked
intrathecal load pressure (Fig. 8). The method also permitted testing of the sensors to detect
certain system-level faults that could occur during deployment. Three variants of the test
structure were used. Tests of the complete system used absolute pressure sensors at the inlet and
outlet ports (Free scale MPXA4250 at each inlet; Free scale MPXA6115 at the outlet) in addition
to the readings available from the manifold pressure sensors. A second variant replaced the dual
polymer reservoirs with a single inlet that delivered liquid under steady pressure from a nitrogen
canister using feedback and set-point control. In this scheme, only one side of the manifold was
used for testing. The second valve in the manifold was flooded with IPA (to remove all air
bubbles from the device) and was sealed at the inlet to allow no air in or alcohol out. The inlet
pressure sensor was a Baratron absolute pressure sensor from MKS. The third variant used the
complete system schematic, except that, instead of using two pressurized reservoirs, one channel
was sealed and only one reservoir was used. This variant was used to test individual channels via
superposition before testing the finished system.

DEPT OF ECE
KADAMMANITTA

13

MZC

SEMINAR 2014

ALPR

Fig. 8. Schematic of the entire system under test. A cylinder at the outlet of the device is used to
simulate intrathecal pressure. The schematic is labeled to signify important nodes at which
partial system test results may be monitored.
.

4.3. SYSTEM-LEVEL DELIVERY TESTS

Typically, the intrathecal pressure for cerebrospinal fluid varies from 0.7 to 1.8 kPa in healthy
adults [28]. This pressure load was mimicked in the experimental procedure by locating the
outlet of the delivery catheter at the bottom of a column of water. The level in the column was
varied to reflect the ranges of pressure differences that can occur in the human body. Detection
of, and compensation for, this pressure variation can be achieved by using the outlet pressure
sensor of the manifold.Experimental characterization of the manifold outlet pressure sensor at a
fixed valve aperture provided a linear relationship between the sensor output and the load
pressure of 35.8 mV/kPa. Further details are available in [27].This information was used,
together with the data from flow characterization of the manifold, to create a microcontroller
program that compensated the aperture to maintain a set flow rate against variation in load
pressure. Fig. 9 compares experimentally the observed flow rates that compensate for variations
in delivery pressure to the uncompensated flow rates. (These data were acquired using the
second variant of the test setup.) The uncompensated flow rates decreased from 0.58 down to

DEPT OF ECE
KADAMMANITTA

14

MZC

SEMINAR 2014

ALPR

0.21 mL/h as the load pressure was increased by 1.5 kPa. In contrast, the compensated flow
program adjusted the manifold to maintain a flow rate of 0.58 mL/h while remaining within
0.5% across the range. In controlled flow tests, both reservoirs, pressurized by the springs, were
filled via the refill ports and used for delivery through the manifold. The flow rate through the
manifold was modulated by varying the duty cycle of a 060-V square wave running at 0.02 Hz.
In a typical test (Fig. 10), three target flow rates were programmed over a 200-min period, and
the resultant output was measured. (These data were acquired using the third variant of the test
setup.) The instantaneous deviation from the set flow rate was no more than 0.01 mL/h. All flow
rates were well within medical limits, with less than 15% deviation from the target delivery rate
[29]. The average supply voltage to the system was 3.3025 V, and the average current draw was
18 mA. Neglecting the power draw from the electronics, this suggests a typical unoptimized
power draw of 57.6 mWh for the system during operation.

Fig. 9. Flow and actuation voltages for changing heights of cylinder pressures to represent
unregulated and regulated flows using the internal pressure sensor. Unregulated flow
varied from 0.58 to 0.21 mL/h, while regulated flow remained within 0.5% of the target
flow rate. The results suggest that the outlet pressure sensor can be used to compensate
to maintain delivery across intrathecal pressure variation.
DEPT OF ECE
KADAMMANITTA

15

MZC

SEMINAR 2014

ALPR

Fig. 10. Typical duty-cycle regulation of flow through the manifold with multiple set points. In
this instance, the duty cycle of a 60-V square wave at 0.02 Hz is altered to achieve mixed
flow at the target delivery rates. Controllable delivery from multiple reservoirs was
achieved with performance acceptable for clinical applications.

DEPT OF ECE
KADAMMANITTA

16

MZC

SEMINAR 2014

ALPR

In the sliding concentric windows (SCW) method was proposed. In this method, license plates
are viewed as irregularities in the texture of the image. Therefore, the abrupt changes in the local
characteristics are the potential license plate. A license plate detection method based on sliding
concentric windows and histogram was proposed. Image transformations are also widely used in
license plate extraction. Gabor filters are one of the major tools for texture analysis .This
technique has the advantage of analyzing texture in unlimited orientations and scales. The result
is 98% when applied to images acquired in a fixed and specific angle. However, this method is
time-consuming. Spatial frequency is identified by using discrete Fourier transform (DFT)
because it produces harmonics that are detected in the spectrum analysis. The DFT is used in a
row-wise fashion to detect the horizontal position of the plate and in a column-wise fashion to
detect the vertical position. The wavelet transform (WT)-based method isused for the extraction
of license plates. In WT, there are four sub ands. The sub image HL describes the vertical edge
information and LH describes the horizontal one. The maximum change in horizontal edges is
determined by scanning the LH image and is identified by a reference line. Vertical edges are
projected horizontally below this line to determine the position based on the maximum
projection. In the HL sub band is used to search the features of license plate and then to verify
the features by checking if in the LH sub band there exists a horizontal line around the feature or
not. The execution time of license plate localization is less than 0.2 s with an accuracy of
97.33%. Adaptive boosting (AdaBoost) is combined with Haar-like features to obtain cascade
classifiers for license plate extraction. The Haar-like features are commonly used for object
detection. Using the Haar-like features makes the classifier invariant to the brightness, color,
size, and position of license plates. The cascade classifiers use global statistics, known as
gradient density, in the first layer and then Haar-like features. Detection rate in this paper reached
93.5%. AdaBoost is also used in . The method presented a detection rate of 99% using images of
different formats, size, and under various lighting conditions. All the methods based on texture
have the advantage of detecting the license plate even if its boundary is deformed. However,
these methods are computationally complex, especially when there are many edges, as in the
case of a complex background or under different illumination conditions.
2.2.4 EXTRACTION USING COLOR FEATURES
DEPT OF ECE
KADAMMANITTA

17

MZC

SEMINAR 2014

ALPR

Since some countries have specific colors for their license plates, some reported work
involves the extraction of license plates by locating their colors in the image. The basic idea is
that the color combination of a plate and characters is unique, and this combination occurs
almost only in a plate region According to the specific formats of Chinese license plates, Shi et
al. proposed that all the pixels in the input image are classified using the hue, lightness, and
saturation (HLS) color model into 13 categories.
A neural network is used to classify the color of each pixel after converting the RGB image
into HLS. Neural network outputs, green, red, and white are the license plate colors in Korea.
The same license plate color is projected vertically and horizontally to determine the highest
color density region that is the license plate region. Since only four colors (white, black, red, and
green)are utilized in the license plates, the color edge detector focuses only on three kinds of
edges (i.e., blackwhite, red white, and greenwhite edges). In the experiment, 1088 images
taken from various scenes and under different conditions are employed. The license plate
localization rate is 97.9% Genetic algorithm (GA) is used as a search method for identifying the
license plate color. From training pictures with different lighting conditions, a GA is used to
determine the upper and lower thresholds for the plate color. The relation between the average
brightness and these thresholds is described through a special function. For any input picture the
average brightness is determined first, and then from this function the lower and upper thresholds
are obtained. Any pixel with a value between these thresholds is labeled. If the connectivity of
the labeled pixels is rectangular with the same aspect ratio of the license plate, the region is
considered as the plate region.
Gaussian weighted histogram intersection is used to detect the license plate by matching its
color. To overcome the various illumination conditions that affect the color level, conventional
HI is modified by using Gaussian function. The weight that describes the contribution of a set of
similar colors is used to match a predefined color. The collocation of license plate color and
characters color is used in to generate an edge image. The image is scanned horizontally and if
any pixel that has a value within the license plate color range is found, the color value of its
horizontal neighbors is checked. If two or more neighbors are within the same character color
DEPT OF ECE
KADAMMANITTA

18

MZC

SEMINAR 2014

ALPR

range, this pixel is identified as an edge pixel in a new edge image. All edges in the new image
are analyzed to find candidate license plate regions. Color images are segmented by the mean
shift algorithm into candidate regions and subsequently classified as a plate or not. The detection
rate of 97.6% was obtained.
A fast mean shift method was proposed. To deal with the problem of illumination variation
associated with the color-based method, proposed a fuzzy logic based method. The hue,
saturation, and value (HSV) color space is employed. Three components of the HSV are first
mapped to fuzzy sets according to different membership functions. The fuzzy classification
function is then described by the fusion of three weighted membership degrees.
2.2.5 EXTRACTION USING CHARACTER FEATURES

License plate extraction methods based on locating its characters have also been
proposed. These methods examine the image for the presence of characters. If the characters are
found, their region is extracted as the license plate region. In some cases instead of using
properties of the license plate directly, the algorithm tries to find all character-like regions in the
image. This is achieved by using a region-based approach. Regions are enumerated and classified
using a neural network. If a linear combination of character-like regions is found, the presence of
a whole license plate is assumed.
The approach used is to horizontally scan the image, looking for repeating contrast
changes on a scale of 15 pixels or more. It assumes that the contrast between the characters and
the background is sufficiently good and there are at least three to four characters whose
minimum vertical size is 15 pixels. A differential gradient edge detection approach is made and
99% accuracy was achieved in outdoor conditions.
Binary objects that have the same aspect ratio as characters and more than 30 pixels are
labeled. The Hough transform is applied on the upper side of these labeled objects to detect
straight lines. The same happens on the lower part of these connected objects. If two straight
DEPT OF ECE
KADAMMANITTA

19

MZC

SEMINAR 2014

ALPR

lines are parallel within a certain range and the number of the connected objects between them is
similar to the characters, the area between them is considered as the license plate area.
In some cases

the characters are extracted using scale-space analysis. The method

extracts large-size blob-type figures that consist of smaller line-type figures as character
candidates. The character region is first recognized by identifying the character width and the
difference between the background and the character region. The license plate is then extracted
by finding the inter-character distance in the plate region. This method yielded an extraction rate
of 99.5%.
In some cases , an initial set of possible character regions are obtained by the first stage
classifier and then passed to the second stage classifier to reject non character regions. Thirtysix
Ada Boost classifiers serve as the first stage classifier. In the second stage, a support vector
machine (SVM) trained on scale-invariant feature transform (SIFT) descriptors is employed. In
some cases, maximally stable extremal regions are used to obtain a set of character regions.
Highly unlike regions are removed with a simplistic heuristic-based filter. The remaining regions
with sufficient positively classified SIFT key points are retained as likely license plate regions.
These methods of extracting characters from the binary image as defining the license plate region
are time consuming because they process all binary objects. Moreover, these methods produce
errors when there is other text in the image.
2.2.6 EXTRACTION COMBINING TWO OR MORE FEATURES

In order to effectively detect the license plate, many methods search two or more features
of the license plate. The extraction methods in this case are called hybrid extraction methods
.Color feature and texture feature are combined in some cases. Fuzzy rules are used to extract
texture feature and yellow colors. The yellow color values, obtained from sample images, are
used to train the fuzzy classifier of the color feature. The fuzzy classifier of the texture is trained
based on the color change between characters and license plate background. For any input image,
each pixel is classified if it belongs to the license plate based on the generated fuzzy rules.
DEPT OF ECE
KADAMMANITTA

20

MZC

SEMINAR 2014

ALPR

In some cases, two neural networks are used to detect texture feature and color feature.
One is trained for color detection and the other is trained for texture detection using the number
of edges inside the plate area. The outputs of both neural networks are combined to find
candidate regions. In some cases, only one neural network is used to scan the image by using H
W window, similar to the license plate size, and to detect color and edges inside this window to
decide if it is a candidate. The neural network is used to scan the HLS image horizontally using a
1M window where M is approximately the license plate width, and vertically using an N 1
window where N is the license plate height. The hue value for each pixel is used to represent the
color information and the intensity is to represent the texture information. The output of both the
vertical and the horizontal scan is combined to find candidate regions.
Time-delay neural network (TDNN) is implemented, to extract plates. Two TDNNs are
used for analyzing color and texture of the license plate by examining small windows of vertical
and horizontal cross sections of the image. The edge and the color information are combined to
extract the plate. High edge density areas are considered as plate if their pixel values are the
same as the license plate. The statistical and the spatial information of the license plate is
extracted using the covariance matrix. The single covariance matrix extracted from a region has
enough information to match the region in different views. A neural network trained on the
covariance matrix of license plate and non license plate regions is used to detect the license plate.
The rectangle shape feature, the texture feature, and the color feature are combined to extract the
license plate. 1176 images that were taken from various scenes and conditions are used. The
success rate is 97.3%. The raster scan video is used as input with low memory utilization. Gabor
filter, threshold, and connected component labeling are used to obtain plate region. Wavelet
transform is used to detect edges of the image. After the edge detection, the morphology in image
is used to analyze the shape and the structure of the image to strengthen the structure to locate
the license plate. A method applies HL sub and feature of 2-D DWT twice to significantly
highlight the vertical edges of license plates and suppress the background noise. Then, promising
candidates of license plates are extracted by first-order local recursive Otsus segmentation and
orthogonal projection histogram analysis. The most probable candidate is selected by edge
density verification and aspect ratio constraint. The license plate is detected using local structure
patterns computed from the modified census transform. Then, two-part post processing is used to
minimize false positive rates.
DEPT OF ECE
KADAMMANITTA

21

MZC

SEMINAR 2014

ALPR

One is the position-based method that uses the positional relation between a license plate
and a possible false positive with similar local structure patterns, such as headlights or radiators.
The other is the color-based method that uses the known color information of license plates.
Reference proposed a method using wavelet analysis and improved HLS color decomposition
and Hough line detection.

Table 2.1 Pros and Cons of Each Class of License Plate Extraction Methods

Methods
Using boundary features

Using global image features

Using texture features

Using color features

Using character features

DEPT OF ECE
KADAMMANITTA

Rationale

Pros

The boundary of license

plate is rectangular.

Simplest, fast and

straightforward

Find a connected object

whose dimension is like a
license plate.
Frequent color transition on
license plate.

Straightforward, independent
of the license
plate position.
Be able to detect even
if the boundary is deformed.

Specific color on license

plate.

Be able to detect inclined

and deformed
license plates
Robust to rotation.

There must be characters on

the license plate.

22

MZC

SEMINAR 2014

Using two or more features

ALPR

Combining features is more

effective.

More reliable.

2.3 LICENSE PLATE SEGMENTATION

The isolated license plate is then segmented to extract thecharacters for recognition. An
extracted license plate fromthe previous stage may have some problems, such as tilt and
nonuniform brightness. The segmentation algorithms should overcome all of these problems in a
preprocessing step. Thebilinear transformation is used to map the tilted extracted license plate to
a straight rectangle.A least-squares method is used to treat horizontal tilt and vertical tilt in
license plate images.
AccordingtoKarhunenLoeve transform, the coordinates of characters are arranged into a
2-D covariance matrix. The eigenvector and the rotation angle are computed in turn. Then,
image horizontal tilt correction is performed. For vertical tilt correction, three methods K-L
transform, the line fitting based on K-means clustering, and the line fitting based on least squares
are put forward to compute the vertical tilt angle . A line fitting method based on the leastsquares fitting with perpendicular offsets was introduced for correcting a license plate tilt in the
horizontal direction. Tilt correction in the vertical direction by minimizing the variance of
coordinates of the projection points was proposed. Character segmentation is performed after
DEPT OF ECE
KADAMMANITTA

23

MZC

SEMINAR 2014

ALPR

horizontal correction and character points are projected along the vertical direction after shear
transform. Choosing an inappropriate threshold for the binarizationof the extracted license plate
results in joined characters. These characters make the segmentation very difficult. License
plates with a surrounding frame are also difficult to segment since after binarization, some
characters may bejoined with the frame. Enhancing the image quality before binarization helps in
choosing the appropriate threshold. Techniques commonly used to enhance the license plate
image are noise removal, histogram equalization, and contrast enhancement. Asystem was
proposed to conduct gradient analysis on the whole image to detect the license plate and then the
detected license plate is enhanced by grey level transformation. A method to enhance only the
characters and to reduce the noise was proposed. The size of the characters is considered to be
approximately 20% of the license plate size. First, the grey-scale level is scaled to 0100, then
the largest 20% pixels are multiplied by 2.55.
Only characters are enhanced while noise pixels are reduced.Sincebinarization with one
global threshold cannot alwaysproduce acceptable results, adaptive local binarization methods
are normally used. Localthresholding is used for each pixel. The threshold is computed by
subtracting a constant c from the mean grey level in an m n window centered at the pixel.
Thethreshold is given by the Niblackbinarization formula to vary the threshold over the image,
based on the local mean and the standard deviation. In the following, we categorize the existing
license plate segmentation methods based on the features they used.
2.3.1 SEGMENTATION USING PIXEL CONNECTIVITY
Segmentation is performed by labeling the connected pixels in the binary license plate
image. The labeled pixels are analyzed and those which have the same size and aspect ratio of
the characters are considered as license plate characters. This method fails to extract all the
characters when there are joined or broken characters.
2.3.2 SEGMENTATION USING PROJECTION PROFILES
Since characters and license plate backgrounds have different colors, they have opposite
binary values in the binary image. Therefore, some proposed methods as in project the binary
DEPT OF ECE
KADAMMANITTA

24

MZC

SEMINAR 2014

ALPR

extracted license plate vertically to determine the starting and the ending positions of the
characters, and then project the extracted characters horizontally to extract each character alone.
In some cases, along with noise removal and character sequence analysis, vertical projection is
used to extract the characters.
By examining more than 30 000 images, this method reachedthe accuracy rate of 99.2%
with a 1020 ms processing speed.character color information is used in the projection instead of
using the binary license plate. By reviewing the literature, it is evident that the method that
exploits vertical and horizontal projections of the pixels is the most common and simplest one.
The pro of the projection method is that the extraction of characters is independent of their
positions. The license plate can be slightly rotated. However, it depends on the image quality.

2.3.3SEGMENTATION USING KNOWLEDGE OF CHARACTERS

Prior knowledge of characters can help the segmentationof the license plate. The binary
image is scannedby a horizontal line to find the starting and ending positionsof the characters.
When the ratio between characters pixelsto background pixels in this line exceeds a certain
thresholdafter being lower than this threshold, this is considered as thestarting position of the
characters. The opposite is done to findthe ending position of the characters.
The extracted license plate is resized into a known template size. In this template, all
character positions are known. After resizing, the same positions are extracted to bethe
characters. This method has the advantage of simplicity. However, in the case of any shift in the
extracted license plate, the extraction results in background instead of characters. The proposed
approach provides a solution for the vehicle license plates that are degraded severely. Color
collocation is used to locate the license plate in the image. Dimensions of each character are used
to segment the character. The layout of the Chinese license plate is used to construct a classifier
for recognition.
The license plates in Taiwan are all in the same colordistribution i.e., black characters and
white background. If the license plate is scanned with a horizontal line, the number of black to
white (or white to black) transitions is at least 6 and at most 14. Hough transform is used to
DEPT OF ECE
KADAMMANITTA

25

MZC

SEMINAR 2014

ALPR

correct the rotation problem, the hybrid binarization technique is used to segment the characters
in the dirty license plate, and feedback self-learning procedure is employed to adjust the
parameters. In the experiment, 332 different images are usedcaptured under various illuminations
and at different distances. The overall location and segmentation rates are 97.1% and 96.4%.

2.3.4 SEGMENTATION USING CHARACTER CONTOURS

Contour modeling is also employed for character segmentation. A shape driven active
contour model is established,which utilizes a variational fast marching algorithm. The system
works in two steps. First, rough location of each character is found by an ordinary fast marching
techniquecombined with a gradient-dependent and curvaturedependent speed function. Then, the
exact boundaries are obtained by a special fast marching method.

2.3.5 SEGMENTATION USING COMBINED FEATURES

In order to efficiently segment the license plate, two ormore features of the characters can
be used. Anadaptive morphology based segmentation approach for seriouslydegraded plate
images was proposed. An algorithm based on the histogram detects fragments and merges these
fragments. A morphological thickening algorithm locates reference lines for separating the
overlapped characters. A morphological thinning algorithm and the segmentation cost calculation
determine the baseline for segmenting the connected characters. For 1189 degraded images, the
entire character content is correctly segmented in 1005 of them.

DEPT OF ECE
KADAMMANITTA

26

MZC

SEMINAR 2014

ALPR

A method was described for segmenting themain numeric characters on a license plate by
introducing dynamic programming (DP). The proposed method functions very rapidly by
applying the bottom-up approach of the DP algorithm and also robustly by minimizing the use of
environment-dependent features such as color and edges. The success rate for detection of four
main numbersis 97.14%.

Methods

Pros

Cons

Using pixelconnectivity .

Simple and straightforward,

robust to the license plate
rotation.

Fails to extract all the

characters when there are
joined or broken characters

Using projection profiles

Independent of character
positions,
be able to deal with some

Noise affects the projection

value, requires prior
knowledge of the number of
license plate characters.

rotation.

Using prior knowledge of

Simple.

Limited by the prior

knowledge, any change may
result in errors

Can get exact character

boundaries.

Slow and may generate

incomplete or distorted
contour.

More reliable.

Computationally complex.

characters

Using character contours

Using combined features

Table 2.2Pros and Cons of Each Class of License Plate Segmentation Methods

DEPT OF ECE
KADAMMANITTA

27

MZC

SEMINAR 2014

ALPR

2.4 CHARACTER RECOGNITION

The extracted characters are then recognized and the outputis the license plate number.
Character recognition in ALPR systems may have some difficulties. Due to the camera
zoomfactor, the extracted characters do not have the same size and the same thickness .Resizing
the characters into one size before recognition helps overcome this problem. The characters font
is not the same all the time since different countries license plates use different fonts. Extracted
characters may have some noise or they may be broken. The extracted characters may also be
tilted.In the following, we categorize the existing character recognition methods based on the
features they used.

2.4.1 CHARACTER RECOGNITION USING RAW DATA

Template matching is a simple and straightforward method in recognitio. The similarity between
a characterand the templates is measured. The template that is the most similar to the character is
recognized as the target. Most template matching methods use binary images because the greyscale is changed due to any change in the lighting.

DEPT OF ECE
KADAMMANITTA

28

MZC

SEMINAR 2014

ALPR

Template matching is performed after resizing the extracted character into the same
size.Several similarity measuring techniques are defined in the literature. Some of them are
Mahalanobis distance and the Bayes decision technique, Jaccard value, Hausdorffdistance, and
the Hamming distance Character recognition uses normalized cross correlation to match the
extracted characters with the templates. Each template scans the character column by column to
calculate the normalized cross correlation. The template with the maximum value is the most
similar one.
Template matching is useful for recognizing single-font,nonrotated, nonbroken, and
fixed-size characters. If a character is different from the template due to any font change,
rotation, or noise, the template matching produces incorrect recognition .

2.4.2 CHARACTER RECOGNITION USING EXTRACTED FEATURES

Since all character pixels do not have the same importanceindistinguishing the character,
a feature extraction techniquethat extracts some features from the character is a good alternative
to the grey-level template matching technique. It reduces the processing time for template
matching because not all pixels are involved. It also overcomes template matching problems if
the features are strong enough to distinguish characters under any distortion. The extracted
features form a feature vector which is compared with the pre-stored feature vectors to measure
the similarity.
The feature vector is generated byprojecting the binary character horizontally and
vertically. In each projection is quantized into four levels. The feature vector is generated from
the Hotelling transform of each character. The Hotelling transform is very sensitive to the
segmentation result. The feature vector is generated by dividing the binary character into blocks
of 33 pixels. Then, the number of black pixels in each block is counted. The feature vector is
generated by dividing the binary character after a thinning operation into 3 3 blocks and
counting the number of elements that have 0, 45, 90, and 135 inclination. The character is
scanned along a central axis. This central axis is the connection between the upper bound
DEPT OF ECE
KADAMMANITTA

29

MZC

SEMINAR 2014

ALPR

horizontal central moment and lower bound horizontal central moment. Then the number of
transitions from character to background and spacing between them form a feature vector for
each character. This method is invariant to the rotation of the character because the same feature
vector is generated. In the feature vector is generated by sampling the character contour all
around. The resulted waveform is quantized into the feature vector. This method recognizes
multifont and multisizecharacters since the contour of the character is not affected by any font or
size change. The Gabor filter is used for feature extraction. The character edges whose
orientation has the same angle as the filter will have the maximum respond to the filter. This can
be used to form feature vector for each character. Kirsch edge detection is applied on the
character image in different directions to extract features.
Using Kirsch edge detection for feature extraction andrecognition achieved better results
than other edge detection methods, such as Prewitt, Frei Chen, and Wallis. In The feature vector
is extracted from the binary character image by performing thinning operation and then
converting the direction of the character strokes into one code. Pixels grey-scale values of 11
subblocks as the features are fed into a neural network classifier. A scene is processed by visiting
nonoverlapping 55 blocks, processing the surrounding image data to extract spread edge
features based on the research conducted , and classifying this subimage according to the coarseto-fine search strategy described. Three character features contourcrossingcounts, directional
counts, and peripheral background area are used. The classification is realized by a support
vector machine. The topological features of characters the number of holes, endpoints, threeway nodes, and fourwaynodesare used. These features are invariant to spatial transformations.

DEPT OF ECE
KADAMMANITTA

30

MZC

SEMINAR 2014

ALPR

3.ADVANTAGES
1. ALPR system quickly and successfully process license plates under different
environmental conditions
2. ALPR system should have the ability to dealwithmultistyleplates.

DEPT OF ECE
KADAMMANITTA

31

MZC

SEMINAR 2014

ALPR

4.FUTURE SCOPE
1. The future research of ALPR should concentrate on multistyle plate recognition, videobased ALPR using temporal information, multi plates processing, high definition plate
image processing, and ambiguous character recognition.
2. System that will help in recognizing the number plates automatically and store them into
a database. Thus we can say it provides a highly secure environment with lower cost by
making the system automatic.
3. For end users it means easy transit through the toll gates and avoiding repeated and hectic
security checks at mall parking entrance and society gates.

5.CONCLUSION

DEPT OF ECE
KADAMMANITTA

32

MZC

SEMINAR 2014

ALPR

This paper is about building a system that will help in recognizing the number plates
automatically and store theminto a database. Thus we can say it provides a highly secure
environment with lower cost by making the system automatic. Automatic vehicle identification is
an essential stage in intelligent traffic systems. In general, an ALPR system consists of four
processing stages. In the image acquisition stage, some points have to be considered when
choosing the ALPR system camera, such as the camera resolution and the shutter speed. In
thelicense plate extraction stage, the license plate is extracted based on some features such as the
colour, the boundary, or the existence of the characters. In the license plate segmentation stage,
the characters are extracted by projecting their colour information, by labelling them, or by
matching their positions with template. Finally, the characters are recognized in the character
recognition stage by template matching, or by classifiers such as neural networks and fuzzy
classifiers. Automatic license plate recognition is quite challenging due to the different license
plate formats and the varying environmental conditions. There are numerous ALPR techniques
have been proposed in recent years. For end users it means easy transit through the toll gates and
avoiding repeated and hectic security checks at mall parking entrance and society gates. In this
project we have presented the review of image processing techniques for license plate
recognition with various approaches.

6. REFERENCE

DEPT OF ECE
KADAMMANITTA

33

MZC

SEMINAR 2014

ALPR

[1] Lekhana G.C, M.Tech; R.Srikantaswamy, Professor, International Journal of Advanced

Technology & Engineering Research (IJATER) Volume 2, Issue 4, July 2012.
[2] Kumar Parasuraman, An Efficient Method for Indian Vehicle License Plate Extraction and
Character Segmentation IEEE International Conference on Computational Intelligence and
Computing Research 2010.
[3] blocksim.reliasoft.com
[4] StutiAsthana, Niresh Sharma, RajdeepSingh,Vehicle number plate recognition using
multiplelayer back
propagation neural networks International Journal of Computer Technology and Electronics
Engineering (IJCTEE)Volume 1, Issue 1
[5] Liannwen JIN, JianZhao QI, "Car plate number characters recognition using Gabor
Orientation features and neural networks.", IEEE International Conference On Neural Networks
and Signal Processing, pp. 1628-1631, 2003

DEPT OF ECE
KADAMMANITTA

34

MZC