Currency Recognition on Mobile Phones

Proposed system modules

Segmentation

Feature Extraction

Instance Retrieval
1. Building a Visual Vocabulary
2. Image Indexing Using Text Retrieval Methods
3. Retrieval Stage
4. Spatial re-ranking
5. Classification

Adaptation to Mobile

Performance analysis

Module description
A. Segmentation
The images might be captured in a wide variety of environments, in terms of lighting
condition and background while the bill in the image itself could be deformed. Image
segmentation is important not just for reducing the data to process but also for reducing
irrelevant features (background region) that would affect the decision-making. This work
starts with a fixed rectangular region of interest (ROI) which is forty pixels smaller from all
four sides than the image itself. This work assumes that a major part of the bill will be present
inside this region. Everything outside this ROI is a probable background. Once this region is
obtained, it must be extended to a segmentation of the entire object. Let x be an image and
let y be a partition of the image into foreground (object) and background components. Let x i
R3 be the color of the ith pixel and let y be equal to +1 if the pixel belongs to the object
i

and to -1, otherwise. For segmentation this work use a graph cut based energy minimization
formulation. The cost function is given by
E ( x , y )= log p ( y i|x i) +
i

S ( y i , y jx)

(i , j)

The edge system E determines the pixel neighborhoods and is the popular eight-way
connection. The pair wise potential S(yi , yj|x) favors neighbor pixels with similar color to
have the same label. Then the segmentation is defined as the minimize arg min y E(x,y). We
use the Grab Cut algorithm, which is based on iterative graph cuts, to carry out foreground/
background segmentation of the images captured by the user. The system should be able to
segment the foreground object correctly and quickly without any user interaction. Whenever
the foreground area is smaller than a pre-decided threshold, a fixed central region of the
image is marked as foreground.
B. Instance Retrieval
5.3.1. Building a Visual Vocabulary
This work first locates keypoints in the foreground region of the image (obtained from
segmentation) and describes the key point regions, using any descriptor extractor like SIFT,
SURF or ORB-FREAK . This work obtains a set of clusters of features using hierarchical Kmeans algorithm. The distance function between two descriptors x1 and x2 is given by
d ( x1 , x2 ) = ( x 1x 2) 1( x1 x2 )

Where is the covariance matrix of descriptors. As is standard, the descriptor space is affine
transformed by the square root of so that Euclidean distance may be used. The set of
clusters forms the visual vocabulary of image.
5.3.2. Image Indexing Using Text Retrieval Methods
For every training image, after matching each descriptor to its nearest cluster, we get a vector
of frequencies (histogram) of visual words in the image. Instead of directly using visual word
frequencies for indexing, we employ a standard term frequency - inverse document
frequency (tf-idf ) weighting. Suppose there is a vocabulary of k words, then each image is

represented by a k-vector

V d =(t 1 , , t i , , t k )

, of weighted word frequencies with

components
ti =

nid
N
log
nd
ni

( )

Here nid is the number of occurrences of word i in document d, n d is the total number of
words in the document d, ni is the total number of occurrences of term i in the whole database
and N is the total number of documents in the whole database. The weighting is a product of
nid
two terms: the word frequency ni , and the inverse document frequency log

( Nn )
i

.However, retrieval on this representation is slow and requires lots of memory. This makes it
impractical for applications on mobile phones. Therefore, we use an inverted index for
instance retrieval. The inverted index contains a posting list, where each posting contains the
occurrences information (e.g. frequencies, and positions) for documents that contain the term.
To rank the documents in response to a query, the posting lists for the terms of the query must
be traversed, which can be costly, especially for long posting lists.
5.3.3. Retrieval Stage
At the retrieval stage, this work obtains a histogram of visual words (query vector) for the
test image. Image retrieval is performed by computing the normalized scalar product (cosine
of the angle) between the query vector and all tf-idf weighted histograms in the database.
They are then ranked according to decreasing scalar product. This work selects the first 10
images for further processing.
5.3.4. Spatial re-ranking
The Bag of Words (BoW) model fails to incorporate the spatial information into the ranking
of retrieved images. In order to confirm image similarity, this work checks whether the key
points in the test image are in spatial consistency with the retrieved images. This work use the
popular method of geometric verification (GV) by fitting fundamental matrix to find out the
number of key points of the test image that are spatially consistent with those of the retrieved
images.

5.3.5. Classification
In the voting mechanism, each retrieved image adds votes to its image class (type of bill) by
the number of spatially consistent key points it has (computed in the previous step). The class
with the highest vote is declared as the result.
C. Adaptation to Mobile
The recognition model needed for retrieval cannot be used directly on a mobile phone
because of the memory requirement. The system was able to adapt the above solution to a
mobile environment by making very significant reductions in complexity, as much as
possible, without sacrificing the effective accuracy. This allows us to achieve the best
possible performance, given the severe restrictions in various aspects of the pipeline that we
have to contend with.
D. Performance analysis
In this step evaluate the performance metrics such as accuracy, and precision for the proposed
system..

CHAPTER 2
INTRODUCTION
2.1 Computer Imaging
It can be defined a acquisition and processing of visual information by computer. Computer
representation of an image requires the equivalent of many thousands of words of data, so the
massive amount of data required for image is a primary reason for the development of many
sub areas with field of computer imaging, such as image compression and segmentation.
Another important aspect of computer imaging involves the ultimate receiver of visual
information in some case the human visual system and in some cases the human visual
system and in others the computer itself.
Computer imaging can be separate into two primary categories:
1. Computer Vision.
2. Image Processing

Fig 1. Computer Imaging

Historically, the field of image processing grew from electrical engineering as an extension of
the signal processing branch, whereas are the computer science discipline was largely
responsible for developments in computer vision.

2.2 Computer Vision

1. Image Analysis: involves the examination of the image data to facilitate solving vision
problem.
The image analysis process involves two other topics:
Feature Extraction: is the process of acquiring higher level image information, such as
shape or color information.
Pattern Classification: is the act of taking this higher level information and identifying
objects within the image.
Computer vision systems are used in many and various types of environments, such as:
1. Manufacturing Systems
2. Medical Community
3. Law Enforcement
4. Infrared Imaging
5. Satellites Orbiting.
2.3 Image Processing
The major topics within the field of image processing include:
1. Image restoration.
2. Image enhancement.
3. Image compression.
1.Image Restoration
Is the process of taking an image with some known, or estimated degradation, and restoring it
to its original appearance. Image restoration isoften used in the field of photography or
publishing where an image was somehow degraded but needs to be improved before it can be
printed

Fig 2. Image restoration

2. Image Enhancement
Involves taking an image and improving it visually, typically by taking advantages of human
Visual Systems responses. One of the simplest enhancement techniques is to simply stretch
the contrast of an image.
Enhancement methods tend to be problem specific. For example, a method that is used to
enhance satellite images may not suitable for enhancing medical images.
Although enhancement and restoration are similar in aim, to make an image look better. They
differ in how they approach the problem. Restoration method attempt to model the distortion
to the image and reverse the degradation, where enhancement methods use knowledge of the
human visual systems responses to improve an image visually.

Fig 3. Image Enhancement

3.Image Compression
Involves reducing the typically massive amount of data needed to represent an image. This
done by eliminating data that are visually unnecessary and by taking advantage of the
redundancy that is inherent in most images. Image processing systems are used in many and
various types of environments, such as:
1. Medical community
2. Computer Aided Design
3. Virtual Reality
4. Image Processing.

Fig 4. Image Enhancement

2.4 Computer Imaging Systems
Computer imaging systems are comprised of two primary components types, hardware and
software. The hard ware components can be divided into image acquiring sub system
(computer, scanner, and camera) and display devices (monitor, printer).The software allows
us to manipulate the image and perform any desired processing on the image data.
2.5 Digitization
The process of transforming a standard video signal into digital image .This transformation is
necessary because the standard video signal in analog (continuous) form and the computer
requires a digitized or sampled version of that continuous signal. The analog video signal is

turned into a digital image by sampling the continuous signal at affixed rate. The value of the
voltage at each instant is converted into a number that is stored, corresponding to the
brightness of the image at that point. Note that the image brightness of the image at that point
depends on both the intrinsic properties of the object and the lighting conditions in the scene.
2.6. Image Representation
We have seen that the human visual system (HVS) receives an input image as a collection of
spatially distributed light energy; this is form is called an optical image. Optical images are
the type we deal with every day cameras captures them, monitors display them, and we see
them [we know that these optical images are represented as video information in the form of
analog electrical signals and have seen how these are sampled to generate the digital image
I(r , c).
The digital image I (r, c) is represented as a two- dimensional array of data, where each pixel
value corresponds to the brightness of the image at the point (r, c). in linear algebra terms , a
two-dimensional array like our image model I( r, c ) is referred to as a matrix , and one row
( or column) is called a vector.
The image types we will consider are:
1. Binary Image
Binary images are the simplest type of images and can take on two values, typically black
and white, or 0 and 1. These types of images are most frequently in computer vision
application where the only information required for the task is general shapes, or outlines
information. For example, to position a robotics gripper to grasp ) )an object or in
optical character recognition (OCR). Binary images are often created from gray-scale images
via a threshold value is turned white (1), and those below it are turned black (0).

Fig 5. Binary Image

2. Gray Scale Image
Gray _scale images are referred to as monochrome, or one-color image. They contain
brightness information only brightness information only, no color information. The number of
different brightness level available. The typical image contains 8 bit/ pixel (data, which
allows us to have (0- 255) different brightness (gray) levels. The 8 bit representation is
typically due to the fact that the byte, which corresponds to 8-bit of data, is the standard small
unit in the world of digital computer.

Fig 6. Gray Scale Images

3. Color Image
Color image can be modeled as three band monochrome image data, where each band of the
data corresponds to a different color.

Fig 7. Color Images

The actual information stored in the digital image data is brightness information in each
spectral band. When the image is displayed, the corresponding brightness information is
displayed on the screen by picture elements that emit light energy corresponding to that
particular color.
Typical color images are represented as red, green ,and blue or RGB images .using the 8-bit
monochrome standard as a model , the corresponding color image would have 24 bit/pixel 8
bit for each color bands (red, green and blue ).
2.7. Introduction to the project
Visual object recognition on a mobile phone has many applications. In this paper, we focus
on the problem of recognition of currency bills on a low-end mobile phone. This is an
immediate requirement for the visually impaired individuals. There are around 285 Million
people estimated to be visually impaired worldwide, out of which 39 Million are blind and
246 Million have low vision. The differences in texture or length of currency bills are not
really sufficient for identification by the visually impaired. Moreover, bills are not as easy to
distinguish by touch as coins. Certain unique engravings are printed on the bills of different
currencies but they tend to wear away.
We adopt an approach based on computer vision on mobile devices, and develop an
application that can run on low end smart phones. We consider the bills of Indian National
Rupee (|) as a working example, but the method can be extended to a wide variety of settings.
Our problem is challenging due to multiple reasons. We want all the computations to happen
on the phone itself and this requires appropriate adaptation of the recognition architectures to

a mobile device. Since our application is desired to be usable in a wide variety of

environments (such as in presence of background clutter, folded bills etc.), we need a robust
recognition scheme that can address these challenges. Also, visually impaired users may not
be able to cooperate with the imaging process by realizing the environmental parameters (like
clutter, pose and illumination).
2.8. Problem Definition Of the Project
Working on a mobile platform brings with it a number of unique challenges that need to be
taken care of. Primarily, the restrictions are in the memory, the application size, and the
processing time. Currently, the average size of an iOS application is 23MB, while the RAM
limit for a Windows phone application is 150MB. For an application to run on a mobile
phone without affecting the others, it should not use more than 100MB of storage and 50MB
of RAM. Our application recognizes the bills in two major steps. First we segment the bill
from the clutter. Then we look at the most similar bill in the database. Though both these
problems can be solved with good performance using many state-of-the-art computer vision
algorithms, they are not really mobile friendly. The recognition model and other necessary
information for our application would typically require more than 500MB of storage and
200MB of RAM with a direct implementation.
This exceeds practical limits by a large amount. To be practically useful, the applications
response time should not be more than 4 seconds keeping in mind that the current average
response time is 3.28 seconds.

CHAPTER 3
LITERATURE SURVEY
1) Monitoring of the Rice Cropping System in the Mekong Delta Using
ENVISAT/ASAR Dual Polarization Data-Alexandre Bouvet, Thuy Le Toan, and Nguyen
Lam-Dao, 2009.
Introduction
In recent years, changes in cultural practices have been observed in different regions of the
world. The rice growth region in the Mekong Delta in Vietnam is a good example of changes
from the traditional to modern rice cultivation system in the last ten years. A multiple
cropping system is implemented, increasing the number of crops per year from one or two to
two, three, or even more. Dike infrastructures have been built and intensified after 2000 to
block the flood way into the fields during the flood season so as to allow an additional crop
cycle. Short-cycle rice varieties (80100 days) are planted in order to harvest three crops per
year instead of one or two. Finally, modern water management has been partly introduced in
the last three years, consisting in intermittent drainage between two irrigation operations. For
those changes in cultural practices, the intensity temporal change method for rice mapping
and monitoring needs to be upgraded. In this work, a method using polarization information
is developed and assessed for this purpose. Because of the vertical structure of rice plants, the
difference between HH and VV backscattering is expected to be higher than that of other crop
or land cover types, and through the relation with wave attenuation in the canopy, the ratio of
the HH and VV backscattering coefficients (hereafter called HH/VV) can be related to the
vegetation biomass. A joint analysis of ERS and RADARSAT-1 data , and the modeling of Cband HH and VV revealed that HH is significantly higher than VV, and the difference can
reach 67 dB at the peak growth stage. Based on these findings, HH/VV is potentially a good
classifier for rice monitoring, and methods using HH/VV need to be developed and assessed.
Specifically, in this work, the method is developed using a time series of dual polarization
(HH and VV) ASAR data and tested in the province of An Giang in the Mekong Delta.
Advantages

This promising result shows that methods using SAR data can be timely and cost
effective.

The method is well-suited to regions where fields have multiple crops and shifted
calendars.

Disadvantages

Need to consider the improvement of the method by using HH/VV and the temporal
change of HH and/or VV in the multi date approach.

2) Rice Phenology Estimation With Multitemporal Terrasar-X Data Using Dynamic

System Concepts F. Vicente-Guijalba, T. Martinez-Marin, J.M. Lopez-Sanchez,
Introduction
Precision farming has been an important subject during the last decades. The aim of these
agricultural techniques is to optimize the field-level management regarding to the crop needs,
the environmental impact and the economical competitiveness of the yield. Remote sensing
tools based on SAR have improved coverage and temporal information resolution for these
agricultural practices. Due to its importance in the human diet, rice has been subject of study
in a wide set of remote sensing works. The first studies with SAR were aimed to detect and
classify rice fields. More recent works have demonstrated that by means of a set of PolSAR
variables it is possible to obtain a coarse estimation of the phenological stage in rice fields.
Phenology provides a measure of the biological progress within a crop field and the
estimation of this parameter allows farm managers to plan crop activities in an optimized
way. Based on the previous approaches, where each estimation is obtained for a single
acquisition without using any other information, This work focused the estimation problem
from a dynamic system view. The main objective is to employ the temporal information
provided by the time series of SAR images to infer the phenological stage in a particular field
and date. The estimation approach consists in two main stages: the dynamical model
generation and the estimation itself.
Advantages

The proposed method is able to provide estimation on rice fields based on dual-pol
SAR imagery.

It achieves results with higher resolution ground truth data in order to validate this
methodology.

Disadvantages

Need to study the generation of models for other kind of crops that behaves in a
similar way and try to apply an analogous approach to the phonological estimation.

3) Rice Phenology Monitoring by Means of SAR Polarimetry at X-Band- Juan M.

Lopez-Sanchez, Senior Member, IEEE, Shane R. Cloude, Fellow, IEEE, and J. David
Ballester-Berman,2012.
Introduction
The feasibility of retrieving the phenological stage of rice fields at a particular date by
employing coherent copular dual-pol X-band radar images acquired by the TerraSAR-X
sensor has been investigated in this paper. A set of polarimetric observables that can be
derived from this data type has been studied by using a time series of images gathered during
the whole cultivation period of rice. Among the analyzed parameters, besides backscattering
coefficients and ratios, we have observed clear signatures in the correlation (in magnitude and
phase) between channels in both the linear and Pauli bases, as well as in parameters provided
by target decomposition techniques, like entropy and alpha from the eigenvector
decomposition. A new model-based decomposition providing estimates of a random volume
component plus a polarized contribution has been proposed and employed in interpreting the
radar response of rice. By exploiting the signatures of these observables in terms of the
phenology of rice, a simple approach to estimate the phonological stage from a single pass
has been devised. This approach has been tested with the available data acquired over a site in
Spain, where rice is cultivated, ensuring ground is flooded for the whole cultivation cycle,
and sowing is carried out by randomly spreading the seeds on the flooded ground.
Advantages

The proposed method is simple.

It provide better estimation accuracy

Disadvantages

The main drawback of using dual-pol TerraSAR-X images for this application is their
narrow swath (around 15 km on the ground), which is too small for devising a
monitoring scheme on large-scale rice plantations.

Noise level of the system (NESZ around 19 dB), which may result very close or
even higher than the backscattering from rice fields, especially at the early stages of
the cultivation cycle.

4) A Kalman Filter Based Mtinsar Methodology For Derving 3d Surface Displacement

Evolutions- Hu J. , Ding X.L. , Li Z.W., Zhu J.J. , Sun Q., Zhang L., Omura M., 2012.
Introduction
Multi-temporal InSAR (MTInSAR) have been used widely for studying earth surface
deformations related to many geophysical processes. However, MTInSAR techniques have
been able to measure one-dimensional (1D) surface deformations in the direction of the lineof-sight (LOS) of the radar. As surface deformations are usually three-dimensional, onedimensional observation apparently cannot always fully reflect the actual deformations. In
addition, the temporal resolution of MTInSAR measurements is limited by the satellite orbit
repeat period. The number of SAR satellites has been increasing rapidly in recent years. It is
therefore very desirable to combine the observations from the different SAR satellites and
orbits to derive more comprehensive surface deformation measurements. This work present
a novel new MTInSAR approach for exploiting multi-sensor, multi-track and multi-temporal
interferograms to infer three-dimensional (3D) surface displacements. The proposed approach
is based on Kalman filter that has been widely used for modeling various dynamic processes.
First, the 1D LOS measurements are estimated from multi-sensor, multi-track and multitemporal interferograms. The observation model and state models of the Kalman filter are
then constructed by considering the imaging geometry and temporal correlation. The 3D
surface displacement at all the acquisition times can be estimated based on the models and a
weighting scheme that reflects the noise levels of the observations and the deformations. The
accuracy of the measurements in the north-south directions is low due to the polar orbits of
the current SAR satellites. In order to ensure the accuracy of the results in the up and eastwest directions, we assume that the deformation in the north-south direction is negligible in
the case study carried out for the Los Angeles area. The experiment uses 21 SAR acquisitions
from ENVISAT ascending and descending orbits and PALSAR ascending obits. The results
are compared with GPS measurements in the area.
Advantages

This work can fully utilize the available interferograms

Significantly increase the temporal monitoring frequency.

Disadvantages

The 3D instaneous rate vectors and correspondingly variances are usually difficult to
exactly identify without any priori information.

5) Kalman-Filter-Based Approach for Multisensor, Multitrack, and Multitemporal

InSAR - Jun Hu, Xiao-Li Ding, Zhi-Wei Li, Jian-Jun Zhu, Qian Sun, and Lei Zhang,
Member, IEEE, 2013.
Introduction
Differential interferometric synthetic aperture radar (SAR) (InSAR) (DInSAR) has been
widely used for monitoring ground deformation associated with various geophysical and
engineering processes. However, the applications of DInSAR have been limited by the effects
of temporal and spatial decorrelation, atmospheric artifacts, and the inability of the method
in providing 3-D measurements. Several multitemporal InSAR (MTInSAR) methods have
been developed in recent years to reduce the effects of temporal and spatial decorrelation and
atmospheric artifacts, including the persistent scatterers, the small-baseline (SB) subset, and
the temporarily coherent point. The measurements from the MTInSAR approaches are,
however, 1-D too, i.e., along the line of sight (LOS) of the SAR satellite. When the ground
moves not in this direction only, which is, in fact, the case most of the time, the InSAR
measurements cannot fully reflect the actual deformation. Some efforts have been made to
derive 2-D or 3-D displacement information by combining InSAR measurements from
different orbits or combining InSAR measurements with other types of measurements such as
those from the Global Positioning System (GPS). This work present a Kalman-filter-based
approach for retrieving 3-D surface displacement from multisensor, multitrack, and
multitemporal SAR interferograms. This approach allows InSAR measurements from
different directions to be integrated sequentially as they become available so that hightemporalresolution results can be achieved. The approach is tested with both simulated and
real SAR data sets to verify its performance.
Advantages

The method works well when the measurement noise is low.

The proposed approach can be potentially used to include other measurements, such
as GPS and leveling, in the solutions.

It achieves the improved accuracy

Disadvantages

It becomes unstable when the measurement noise is high due to the polar-orbiting
imaging geometries of the current satellite SAR sensors.

6) Estimating near future regional corn yields by integrating multi-source observations

into a crop growth model -Jing Wang, Xin Li Ling Lu, Feng Fang, 2013.
Introduction
Regional crop yield estimations play important roles in the food security of a society. Crop
growth models can simulate the crop growth process and predict crop yields, but significant
uncertainties can be derived from the input data, model parameters and model structure,
especially when applied at the regional scale. Abundant observational information provides
the relative true value of surface conditions, and this information includes those areal data
from remote sensors and ground observations. The objective of this study was to present a
data fusion framework used to calibrate a crop growth model at the plot scale and to estimate
yield at the regional scale on the basis of two types of data fusion algorithms, which reduces
the uncertainty of regional yield estimations. First, based on local intensive observation, the
simulated annealing algorithm was applied to obtain a parameter vector that was suited to the
local crop variety. This scheme reduces model parameter uncertainty. Then, the ensemble
Kalman filter (EnKF), a sequence filter algorithm, was adopted to integrate the areal crop
growth information that was derived from remote sensing technologies into a crop growth
model for precise regional yield estimation, which reduces uncertainties in the model
structure or input data related to meteorological, soil, or filed management information. This
proposed scheme and technology will provide an operational method for precisely estimating
crop yields at regional scales.
Advantages

The WOFOST model can simulate the growth curve and yield of corn, especially with
respect to crop carbon absorption in agri-ecological systems

This study aimed to assess the feasibility of assimilating areal observation data into a
crop growth model to improve spatial estimates of crop yields and carbon pools.

Disadvantages

Estimation uncertainty also arises from parameter uncertainty, and an accurate

parameter set is critical for accurate yield predictions.

7) Efficient Spatio-temporal Mining of Satellite Image Time Series for Agricultural

Monitoring- Andreea Julea, Nicolas Meger, Christophe Rigotti, Emmanuel Trouve,
Romain Jolivet, and Philippe Bolon, 2012.
Introduction
This work presents an unsupervised technique to support SITS analysis in agricultural
monitoring. The presented approach relies on frequent sequential pattern extraction along the
temporal dimension, combined with a spatial connectivity criterion. It allows to uncover sets
of pixels satisfying two properties of cultivated areas: they are spatially connected/grouped
and share similar temporal evolutions. The approach requires no prior knowledge of the
objects (identified regions) to monitor and needs no user-supplied aggregate functions nor
distance definitions. It is based on the extraction of patterns, called Grouped Frequent
Sequential patterns (GFS-patterns), satisfying a support constraint and a pixel connectivity
constraint. In this paper, we extend the general framework of GFS-patterns. This work
proposed in two directions, when applied to agricultural monitoring. Firstly, we show that,
even though the connectivity constraint does not belong to any typical constraint family (e.g.,
monotonic, anti-monotonic), it can be pushed partially in the search space exploration. This
leads to significant reduction of execution times on real Satellite Image Time Series of
cultivated areas. Secondly, we show that a simple post-processing using a maximality
constraint over the patterns is very effective. Indeed, it restricts the number of patterns to a
human-browsable collection, while still retaining highly meaningful patterns for agromodelling. This property is confirmed even for poor quality inputs (rough image
quantization, raw noisy images).
Advantages

GFS-patterns is used to extract sets of pixels sharing similar evolution from Satellite
Image Time Series over cultivated areas

It achieves reduced GFS-patterns extraction times

Even on poor quality inputs (i.e., noisy images, rough quantization), the method can
exhibit various level of details of primary interest in agro-modelling

Disadvantages

The contribution due to the stratified atmosphere can be roughly estimated by using
DEMs and meteorological data, but the effects of the turbulent atmosphere still
degrade interferograms.

8) Integrating Vegetation Indices Models and Phenological Classification with

Composite SAR and Optical Data for Cereal Yield Estimation in Finland (Part I)Heikki Laurila, Mika Karjalainen, Juha Hyypp and Jouko Kleemola, 2010.
Introduction
The aim of the present study was to estimate actual non-potential grain yield levels for high
latitude spring cereals (spring wheat, barley and oats, Avena Sativa L.) in large area field
conditions in southern Finland. The cereal theoretical maximum yielding capacity is limited
by environmental and vegetation stresses (e.g., drought periods, nutrient deficiencies,
pathogen epidemics) during growing season in actual field growing conditions. These stress
factors result to reduced non-potential baseline yield levels (yb, kg/ha) on field parcel level.
The objectives of the present study were: (i) to construct a dynamic SatPhenClass
phonological classification model, which classifies both optical and SAR satellite data based
on cereal actual phenological development in both vegetative and generative phases (ii) to
calibrate and validate multispectral Composite Vegetation Indices (VGI) models, which
integrate both phenologically preclassified optical (Models III) and microwave SAR data
(Composite SAR and NDVI Model III), and finally (iii) VGI models were used to estimate
cereal non-potential baseline yield (yb) levels in growing zones (IIV) in southern Finland
during 19962006.
Advantages

The proposed method is validated to estimate cereal yield levels using solelyoptical
and SAR satellite data.

The averaged composite SAR modeled grain yield level was 3,750 kg/ha (RMSE =
10.3%, 387 kg/ha) for high latitude spring cereals.

Disadvantages

The early emergence in vegetative phase (ap, BBCH 012) in two leaf stage before
double ridge induction and the senescence phase after full maturity and harvest (dp),
BBCH > 90) were difficult to estimate.

9) Multi-temporal MODISLandsat data fusion for relative radiometric normalization,

gap filling, and prediction of Landsat data David P. Roy, Junchang Ju, Philip Lewis ,
Crystal Schaaf , Feng Gao, Matt Hansen, Erik Lindquist, 2008.
Introduction
A semi-physical fusion approach that uses the MODIS BRDF/Albedo land surface
characterization product and Landsat ETM+ data to predict ETM+ reflectance on the same,
an antecedent, or subsequent date is presented. The method may be used for ETM+
cloud/cloud shadow and SLC-off gap filling and for relative radiometric normalization. It is
demonstrated over three study sites, one in Africa and two in the U.S. (Oregon and Idaho)
that were selected to encompass a range of land cover land use types and temporal variations
in solar illumination, land cover, land use, and phenology. Specifically, the 30 m ETM+
spectral reflectance is predicted for a desired date as the product of observed ETM+
reflectance and the ratio of the 500 m surface reflectance modeled using the MODIS BRDF
spectral model parameters and the sun-sensor geometry on the predicted and observed
Landsat dates. The difference between the predicted and observed ETM+ reflectance
(prediction residual) is compared with the difference between the ETM+ reflectance observed
on the two dates (temporal residual) and with respect to the MODIS BRDF model parameter
quality. For all three scenes, and all but the shortest wavelength band, the mean prediction
residual is smaller than the mean temporal residual, by up to a factor of three. The accuracy is
typically higher at ETM+ pixel locations where the MODIS BRDF model parameters are
derived using the best quality inversions. The method is most accurate for the ETM+ nearinfrared (NIR) band; mean NIR prediction residuals are 9%, 12% and 14% of the mean NIR
scene reflectance of the African, Oregon and Idaho sites respectively.
Advantages

The proposed method Achieves best quality

Also achieves higher accuracy

Disadvantages

Significant reflectance changes of this nature are difficult to accommodate using

conventional relative radiometric normalization and gap filling techniques.

10) An automated algorithm to detect timing of urban conversion of agricultural land

with high temporal frequency MODIS NDVI data - Bhartendu Pandey, Qingling Zhang
and Karen C. Seto, 2013.
Introduction
Urban expansion is one of the major drivers of agricultural lands loss. However, current
remote sensing-based efforts to monitor this process are limited to small scale case studies
that require much user input. Given the rate and magnitude of contemporary urbanization,
there is a need to develop a land change algorithm that can characterize the loss of
agricultural land at large scales over long time periods. Moreover, characterizing agricultural
land conversion trajectories from remote sensing images is complex due to farm size, climatic
variability, changes in cropping patterns, and variations in the rate of development processes.
Here This work propose an econometric time series approach to identify agricultural land
loss due to urban expansion, utilizing high temporal frequency MODIS NDVI data between
2000 and 2010. The algorithm is comprised of two main components: 1) detrending the time
series, and 2) testing for the presence of a breakpoint in the detrended time series and
estimating the date of the breakpoint. Evaluations of the algorithm with simulated and actual
MODIS NDVI data confirm that the method can successfully detect when and where urban
conversions of agricultural lands occur. The algorithm is simple, robust, and highly
automated, thus is valuable for monitoring agricultural land loss at regional and even global
scales
Advantages

The proposed method enables processing of very large datasets, either in spatial extent
or through time

It reduces mistakes due to interpretation or human error.

Disadvantages

This step-wise land-use transitions result into deviation from the assumption of
sequential phases in land conversion process and limit the application of most change
detection algorithms.

CHAPTER 4
CONCLUSION AND FUTURE WORK
CONCLUSION
Visual object recognition is an recent trend which is used to recognize the objects visually
through the systems. Currency recognition through mobile phones will be a most effective
methodology which will be most useful for visually impaired persons. In this work, we have
ported the system to a mobile environment, working around like limited processing power
and memory, while achieving high accuracy and low reporting time. Currency retrieval and
thereafter recognition is an example of fine-grained retrieval of instances which are highly
similar. Thus the result of our experimental results proves that it is more robust to
illumination changes than the SIFT descriptor.
FUTURE WORK
The system implemented in our work is used to implement on Indian currency rupees
whereas in further research it can be implemented to support a world level currency notes.

REFERENCE
1. Jayaguru Panda, Michael S. Brown, C. V. Jawahar, Offline Mobile Instance Retrieval
with a Small Memory Footprint, Computer Vision (ICCV), PP: 1257 1264, 1-8 Dec.
2013
2. A. Ms. Trupti Pathrabe and B. Dr. N. G. Bawane , Paper Currency Recognition System
using Characteristics, International journal latest trends in computing, Vol 1, Issue 2,
December, 2010
3. H. Hassanpour and E. Hallajian, Using Hidden Markov Models for Feature Extraction in
Paper Currency Recognition, Advances in Soft Computing and Its Applications, volume
8266, 2013, pp 403-412
4. Bhawani sharma, Amandeep kaur, Vipan, Recognition of indian paper currency based on
LBP, International journal of Computer applications, Vol 59,Dec 2012
5. Faiz M. Hasanuzzaman, Xiaodong Yang, and YingLi Tian,

Robust and effective

component based bank-note recognition by SURF features, Wireless and Optical

Communications Conference, 15-16 April 2011
6. David M. Chen, Sam S. Tsai, Ramakrishna Vedantham, Radek Grzeszczuk, Bernd Girod,
Streaming Mobile Augmented Reality on Mobile Phones, Mixed and Augmented
Reality, 181 182, 19-22 Oct. 2009
7. Jayaguru Panda, Jayaguru Panda, C.V. Jawahar, Heritage App: Annotating Images on
Mobile Phones, Proceedings of the Eighth Indian Conference on Computer Vision,
Graphics and Image Processing, Article No. 3, 2012
8. Wei Di, Catherine Wah, Anurag Bhardwaj, Robinson Piramuthu, and Neel Sundaresan,
Style Finder: Fine-Grained Clothing Style Recognition and Retrieval, Computer Vision
and Pattern Recognition Workshops, 23-28 June 2013
9. Josef sivic, Andrew zisserman, Video Google: A text retrieval approach to object
matching in videos, Computer Vision, 2003. Proceedings. Ninth IEEE International
Conference on 13-16 Oct. 2003, 1470 - 1477 vol.2

10. Petri Tanskanen, Kalin Kolev, Lorenz Meier, Federico Camposeco, Olivier Saurer, Marc
Pollefeys, Live metric 3D Reconstrction on mobile phones, Computer Vision (ICCV),
2013 IEEE International Conference on 1-8 Dec. 2013