Partial Image Retrieval Systems in Luminance and Color Invariants : An Empirical Study

Integrated Intelligent Research (IIR) International Journal of Web Technology
Volume: 04 Issue: 02 December 2015 Page No.95-99
ISSN: 2278-2389
95
Partial Image Retrieval Systems in Luminance and
Color Invariants : An Empirical Study
P.Mary Jeyanthi
Vels University, Pallavaram, Chennai.
Abstract - Color of the surface is one of the most imperative
characteristics in the process of recognition as well as
classification of the object which is based on camera. On the
other hand, color of the object sometimes differs a lot due to
the difference in illumination as well as the conditions of the
surface. Utilization of the diverse features of the color gets
impeded due to such variations. However, Characterization of
the color of the object is possible with a controlling tool known
as color invariants without considering the factors such as-
illumination and conditions of the surface. In the research
proposal, analysis has been done on the estimation procedure
related to RGB images color invariants. Object color is an
imperative descriptor that can find the corresponding matching
object in applications based on image matching as well as
search, like- Object searching based on template and CBIR
otherwise known as Content Based Image Retrieval. But, many
times it has been observed that the apparent color of different
objects gets varied significantly due to illumination, conditions
of the surface as well as observation (Finlayson et al., 1996).
I. INTRODUCTION
In support of Illumination Invariant indexing method and
characterization of object color, color Invariants or use of
photometric invariants are taken into account. It is because,
object apparent color is dependent on physical process like-
scattering of light at the surface of the object. Apart from that,
the photometric invariant technique has been considered for the
estimation of the components that are regulated by the color of
the object which can be possible by eradicating all the other
components as well as variations and only by taking ray
observation into consideration which is a physical process. For
the Three Dimensional Object Recognition, its shape and
motion assessment as well as correspondence Photometric
Invariants are utilized. Better-quality results can be noticed in
this process as comparison to the utilization of RGB color as
well as the method of grayscale (Healey & Slater,1997). In the
research proposal Geusebroek’s color invariants have been
analyzed on the basis of the Physical Surface Observation
Theory by Kubleka-Munk as well as Gaussian Color Model.
The derived invariants have been utilized in the form of
descriptors instead of grayscale to solve the key point
correspondence. In the analysis, estimation of Geusebroek’s
color invariants by RGB colors has been comprehensively
described and it has been found that it is necessary to use the
estimation of cyclic distance on the invariants’ values.
Utilization of the distance measure is a sign of the same nature
of color corresponding to the invariant spectra’s positive as
well as the negative extremities. The invariant characteristic of
a novel color is known as the definition of the composite
function related to the two fundamental invariants. This
particular invariant is for the improvement of the selectivity
related to the colors of the object with the intention of perfect
matching of similar objects under dissimilar illumination.
Utilization of these two factors- distance measure as well as
invariant is carried out as comparison to other invariants as
well as luminance in different experiments for the keypoint
matching of images along with the process under variable
conditions of illumination of partial image retrieval.
(Lindeberg.,1994).
II. OBJECTIVES
The key objective of the research proposal is about the analysis
of the estimation procedure related to the RGB images’ color
invariants. The major purpose of this research proposal is to
 Put forward an innovative characteristic of color related to
the invariant on the basis of fundamental invariants so that
the feature will meet the circular continuity existing in the
mapping. This mapping is done among the different colors
as well as the correlated invariants (Rao & Ballard,1995).
 Improvement in the performance sector of the image
retrieval in the continuous alteration of illumination
conditions related to the partial object image matching has
been noticed throughout many experiments. These
experiments were carried out with the suggested invariant
with the combined effect of luminance.
 In the first segment of the research proposal the introduced
theory and experiment of Geuse broek related to the
estimation as well as definition of the color invariants have
been analyzed and evaluated. In the subsequent segment
the next objective of the proposal will be executed which
is about the presentation of the two imperative factors-
distance measures and the new feature of the color
invariant. (Slater and Healey,1995).
As mentioned earlier,
 The prime goal of the research proposal is to indicate a
topic of concern that has occurred in due course of object
matching that is based on color invariant by appraising the
Geusebroek’s fundamental color invariants.
 Composite Color invariant H’ has been proposed for the
addressed issues that have taken place due to the use of the
invariants for the evaluation of color matching. The plus
points associated with the application of image retrieval
and the process of color image matching hence gets
evaluated which has become a major part of the objective
(Witkin. Scale-space filtering,1983).
Many prominent image retrieval systems developed by high-
flying commercial organizations as well as educational
institutes and their distinctive characteristics have been
discussed in the thesis. In the last segment, future works in
relation to image retrieval have been discussed thoroughly and
how this process can be used in a practical manner, research
related to this concept has been done.

ISSN: 2278-2389
96
III. PROBLEM STATEMENT
The image retrieval systems generally belong to leading
commercial organizations as well as academic organizations
and therefore, it was quite difficult to get accessed into these
institutes for the research of color invariants and image
matching. From the era of 1990, image retrieval systems has
turned out to be one of the top research areas and most of the
commercial associations and educational foundations have
built such systems that support options like- browsing
arbitrarily or in a random manner, three types of searches that
include search by means of example, sketch and text (key
wording, speech) and lastly navigation with custom-made
images. (Rui, Huang & Chang 1999). These set of options are
available in the image retrieval systems for the execution of
search processes. However, methodical analysis indicates the
implementation of all these options by exploring the facts
related to the trade-offs as more things have to come in the
sector of practical applications which involves real users.
There are numerous visual features that can be explored in case
of applications for computer as well as image retrieve. Multiple
representations are there attached with each of the visual
feature and here the problematic part rises which is about the
perception of human beings which could be different from
different perspective views.
Understanding the high dimensional Indexing is another tough
part in the research proposal as indexing techniques of
multidimensional nature for better efficiency are still needed in
the image retrieval sector which makes the proposal confusing.
It is because more areas are required for exploration so that
challenges like- high dimensionality and Non-Euclidean
Similarity Measure can be solved. There are not ample amount
of research available in the topics such as- the queries related
to the retrieval part, future scope of this process, problems
regarding the larger database of image collection and quick
browsing, searching matching and then retrieval. There is no
given solution regarding the database management related to
the difficulties of the scalability nature which leads to larger
sized collections of images that reduces the efficiency of the
system (Koenderink & van Doorn,1987).
IV. RESEARCH METHODOLOGY
For concrete piece of information about the different image
retrieval systems, it required to get profound knowledge about
the practical applications associated with the modern as well as
conventional systems. For this, exploration and methodical
studies were involved with the proposal with the help of
leading organizations and major universities. The research
methodology comprises the analyses of multiple color digital
images and study of the fundamental characteristics H and C of
the color invariants. The research project mainly focuses on the
estimation of color invariants of the RGB images. For the
future approaches attached with this particular method known
as image retrieval, study of different factors like- color, spatial
relations, illumination and surface conditions, shape etc have
been done. The procedure along with the evaluation part of the
research proposal needs different queries that are significant
for the completion of the evaluation part and for that diverse
images have been experimented though the image retrieval
process under different luminance and different surface
conditions. All the queries are properly analyzed with the help
of professionals of the institutes who have given the profound
knowledge of color variants and image matching as well as
several factors and theories related to the partial image
retrieval process. During the analysis of distinctive
characteristics related to diverse systems present in the
academic institutes as well as in commercial organizations,
with the help of authorization to the organizations, in-depth
knowledge about QBIC, Virage, RetrievalWare, Photobook,
VisualSEEk, WebSEEk, Netra, MARS, Chabot, Informedia
could be achieved (Healey & Slater,1997). The specialists of
image retrieval systems provide the necessary facts about the
image formation as well as invariants; along with it study about
the Gaussian color model was comprehensively instructed with
the invariants (H and C) that are estimated from the RGB
images. The analysis regarding the experiment was thoroughly
discussed with two different issues which are- discontinuity
related to the characteristics and finding the edges of opponent
color with the utilization of θH. Color invariant feature H’ was
properly calculated and that has been mentioned in this
research proposal. Image data, Local descriptor, Local features
Correspondence, Correspondence of regions, matching region
candidates and their correlations, appraisal scores come under
the application part of image matching with clear-cut concept
of retrieval (Finlayson et al., 1996).
Color Invariants
In this particular segment, a comprehensive review regarding
the Geusebroek’s fundamental color variants has been
illustrated. An issue that often occurs while using the basic
color invariants in the object matching has been also given in
this section. Invariants and the Color image formation The
most concerning parts while observation is carried out about
the object color is- the source of light, observer’s point of view
and the object. When a light source illuminates an object, the
reflection spectrum is-
E (λ, x) = e (λ, x) (1−ρf (x))2
R∞ (λ, x) + e (λ, x) ρf
(x)…………………………………………………………... (1)
Where,
x = position within the image
λ = wavelength
e(λ, x) = illumination spectrum of the light
ρf (x) = Fresnel reflectance at x
R∞(λ, x) = material reflectivity ( of the color).
The derived reflectivity of the Kubelka-Munk theory is
originated by the multiplication of reflectivity and illumination
in the equation for the colorants layered objects. Two major
assumptions related to the derivation has been proposed, one is
about the satisfactory thickness of the layer of colorant and the
reflection of the object that can be overlooked on the basis of
colorant and second assumption is about ρfR∞ is almost equal
to zero.
When the positional space which is one dimensional in nature
is taken into consideration along with the standardized spectral
components in white illumination, i(x) can replace e(λ, x).
Here, E(λ, x) = i(x){ρf (x) + (1 − ρf (x))2
R∞(λ,
x)}………………………………………………………….. (2)
And
𝐻 =
𝜕𝐸
𝜕λ
𝜕2𝐸
𝜕λ2
=
𝐸𝜆
𝐸λλ
………………………………………… (3)
From equation 2, assumption regarding matte-surfaced objects
by ignoring Fresnel reflectance, ρf (x) ≈ 0

ISSN: 2278-2389
97
E(λ, x) = i(x)R∞(λ, x)………………....……………………. (4)
Then,
𝐶 =
1
𝐸(λ,x )
𝜕𝐸
𝜕λ
=
𝐸λ
𝐸
………………………………………... (5)
H and C are the features of illumination of invariant color and
these features are used for the characterization of the object
color in different conditions of illumination. In the given two
different illumination conditions, L which signifies luminance
along with H and C which signify color invariants have been
compared. Here in the ALOI database, illumination is used and
from the database the images have been selected. The
illumination used in this experiment has a flat spectrum with a
continuous manner and all these are found in its visible
wavelength. A halogen lamp of OSRAM with tungsten
element has been used which is of 64637 type and under
3100K color temperature. The changes in L are greater than the
variations associated with invariants and visible when
comparison is done with H. Calculation of L is done in relation
to L= 0.299R+0.587G+0.114B with the help of RGB
components.
With the knowledge about color invariants, reviews regarding
the fundamental color invariants by Geusebroek will be easy
and apart from this, the issues related to the color invariant-
based object matching can be avoided. H and C are the
fundamental color invariants which have been discussed in this
proposal with factors like illumination conditions and surface
of the object taken into consideration. The estimated features
are highlighted and illustrated in grayscale along with the
variations linked with illumination. These variations are
highlighted with the help of shadow as well as intensity with
the changes in surface conditions with the help of glossy
surface by Fresnel reflectance.
The Gaussian Color Model
From the digital images, estimation of the invariants are done
with the help of the components of RGB, the Gaussian Color
Model and the conversion related to the RGB is used for it.
If E(λ) is the true energy linked with incident light at λ
wavelength, by using a spectral aperture with a Gaussian
function G(λ; λ0, σλ), then the observation-
Ê λ0 σλ
= ∫ 𝐸(λ) G(λ; λ0, σλ) dλ.
…………………………………………………………….... (6)
Estimated H and C from RGB images
Estimation of the invariant H from RGB components of the
particular image,
H =
0.3𝑅+0.04𝐺−0.35𝐵
0.34𝑅−0.6𝐺+0.17𝐵
…………………………………………………………….. (7)
Now invariant C can be estimated as,
C =
0.3𝑅+0.04𝐺−0.35𝐵
0.06𝑅+ 0.63𝐺+ 0.27𝐵
………………….…………………… (8)
V. ANALYSIS
From equation 7 and 8 the nature of invariants can be
estimated with the linear combinations of R, G and B.
(0.3 – 0.34H)R + (0.04 + 0.6H)G + (-0.35 – 0.17H)B =
0…………………………………………………………..… (9)
And
(0.3 - 0.06C)R + (0.04 – 0.63C)G + (-0.35 – 0.27C)B =
0…………………………………………………………… (10)
From equation 9 and 10 the set points in RGB space which
have similar values of invariant will be at its subspace that is a
two dimensional plane.
Color Invariant Feature H’
Calculation of the color invariant feature H’ can be done by
combining the invariants H with C. Colors on the equal-C
plane Pc along with equal-H plane Ph have been illustrated with
different invariant values θC and θH. The regions with color
illustrate the cross sectional areas of the RGB cube in the form
of Pc and Ph. O is the origin of the RGB space Vector VC and
VH are known as the axes of rotation of the cross sections Pc
and Ph. The two major issues of feature discontinuity and
opponent color edge detection during the evaluation of similar
color can be identified with the analysis based on The Gaussian
Color model, estimation of H and C from RGB images as well
as color invariant feature H’. Therefore, study about these
topics has been proved helpful throughout the whole research
proposal and its framework.
VI. LITERATURE SURVEY
Image Data
The Selection of images that are used in this experiment is
from ALOI which is Amsterdam Library of Object Images.
ALOI comprises a collection of one thousand objects’ images
which have photographed under diverse conditions of
luminance as well as camera angles.
Local Descriptor
Scale-rotation invariant image structure is very much necessary
for the experiment along with stability to any sort of noise.
Thus, implementation of SIFT otherwise known as Scale-
Invariant Feature Transform in the form of local descriptors
related to the illumination variation has been taken into
consideration(Finlayson et al., 1996).
Different image retrieval systems have been analyzed
throughout the research proposal with the general features
present in them.
QBIC: QBIC system has been developed by IBM and the
image retrieval system is concerned with the features like0
color, shape and texture of the images. By the identification of
examples related to query image, different queries are entered
on the screen. The other process of entering query is about the
specification of sketch of a defined shape.
Virage: This is known as VIR Image Engine which has been
developed by Virage Inc. Utilization of several primitives, like-
texture, global color as well as local color and shapes are the
main features of the system.
Photobook: It is a system which has been developed on an
experimental basis by the MIT (Massachusetts Institute of
Technology). With the help of computing shape, textures and
other important features, characterization of object images are
retrieved. Many applications such as- textures, shape and faces
of human are used in the image retrieval technique.
Chabot: It is otherwise known as Cypress. Combined features
of text based and color based are the main features of this
technology.
VisualSEEk: Columbia University has developed this
experimental system. On the basis of spatial location, shape
and color along with the features of keyword image retrieval is
performed.

ISSN: 2278-2389
98
MARS: University of Illinois has designed this system. It is
based on the similar judgment capability of the individuals
which is known as relevant feedback. Quite a few features are
used for the identification of objects with comparison of query
by similarity measures and image storing.
Netra: Image segmentation method is the unique kind of
feature of this system which uses spatial location, shape, color
and texture for the image retrieval.
Informedia: Video retrieval is the specialty of this system with
the integration technology of Image with speech processing.
Color histogram, motion vectors, audio tracks, sound tracks,
speech are used for the identification of video scenes.
RetrievalWare: It is a retrieval engine which is based on
content-based image retrieve which has been developed by
Excalibur Technologies Corp. Special emphasis on the topic of
neural nets has been given here for image retrieval. Color,
shape, brightness, texture, layout, aspect ratio are the features
here.
Application of Image Retrieval Systems
Here in the thesis, for the storage of large-sized data regarding
images, IDBMS has been used. IDBMS means Image Database
Management Systems and this system supports content-based
image retrieval with great efficiency. Similar to the other
Database Management Systems, this IDBMS has also a storage
facility and here all the data gets arranged in relation to
different natures of data models. Query facility is the distinct
feature of it and here images get retrieved in line with the
queries. The queries in this developed system, address the
logical database rather than that of the physical database where
raw data of images get stored. In general, here the selection
queries are utilized.
e.g.- On the basis of the content, selection of images is taken
place. Creating a query facility which can load large size of
images from the database and then check all the images to
observe if it gets fit with the selected criteria, imagination of
this type of idea can generate ground-breaking application in
the sector of image retrieval system. However, this query
facility seems to be inefficient in case of extreme conditions.
Here, the best option is to get all the possible information
related to the images that have been stored in the database and
thereby it can be utilized for the evaluation process of the
selection criterion and it is called indexing. Here in this
particular system indexing of biomedical images, image
storage and image retrieval are three key processes. When the
image data gets stored in the database, an index gets
constructed automatically and stored within the database. At
this point, the information present in the indexes is only useful
for the execution of queries. Only after the conformation
regarding the selection criterion by the index, images can be
retrieved. Therefore, in this particular system, all the
information regarding content that is queried must get index.
Efficient processing in case of the queries is the plus point in
this system because of the modernized techniques of indexing.
The indexing techniques of the traditional DBMSs used in the
image databases of biomedical sector should be replaced with
this highly developed content-based querying of ultramodern
indexing techniques. Here, this system involves two kinds of
indexes Text indexes and content similar indexes. Text indexes
are for the availability of the related queries in time of need
and with this, image retrieval become much easier due to
labeling. But in case of content similar indexes, these indexes
retrieve all the images with similar color of the sample image.
In the development phase of an IDBMS, some additional data
types beyond the model such as- images, arrays are needed. It
is called Object Rational Data Model and in case of this
retrieval system this model is required with the capacity of
storing large images with a nonstop mode.
The key components of the retrieval engine are- Methods of
storage and acess, Visual Feature Extraction, Calculation of the
distance measure similarity and GUI interaction techniques.
Before this technology, text-based searching technique is the
one and only reliable methods for the retrieval of medical
images. Comparison of this modern technology which is based
on content-based retrieval with the older techniques of image
retrieval is given below-It follows some of the key limitations
and these points are considered as the reason for which
conventional methods were eradicated from the image retrieval
process related to the biomedical imaging. Those systems have
manual style of annotations and it takes much time for the
processing than that of the abovementioned retrieval system. It
was also quite expensive to implement in the sector biomedical
fields. When number of images in the database gets increased,
the chances of finding the preferred image also get increased.
Subjective perception and the related difference is another
problem in case of manual annotation. Medical images consist
of the illustrations of numerous objects that indicate specific
information. But, diverse interpretations related to the same
pathological field may be different according to diverse
radiologists and therefore, obtaining profound knowledge
about the concepts, feelings and perceptions related to an
image is quite necessary and this objective gets fulfilled by the
content-based image retrieval system described before.
Illustrative ideas in the format of different shapes are easier
than that of the other forms as biomedical images contain
diverse shapes of organic objects. Therefore, image retrieval
with exactly similar contents makes it simple. Nowadays, low
quality resolution feature along with strong noises are two
major problems in the sector of biomedical imaging. So, this
system has been developed in which higher resolution of
images with less noise facilitates precise segmentation and
feature extraction of the contents.
VII. SCOPE OF THE WORK
In this experimental analysis of joint use of luminance and
color invariants in Partial image retrieval system,
comprehensive knowledge regarding the different nature of
fundamental features H and C of illumination invariants has
been properly illustrated which was introduced by Geusebroek.
The proposal is about the different commercial as well as
academic image retrieval systems and their general features. It
will help to know about the present, past and future
achievements in the technological sector of image retrieval.
Image retrieval system has become a core part in the sectors of
Construction engineering, property protection, Advertising as
well as press, treatment processes in medical science like-
surgeries and treatment of injuries along with other diseases,
research of cultural heritage and many more(Finlayson et al.,
1996). Therefore, profound knowledge about the systematic
approaches related to the image retrieval technique will help
many researchers to gain ample amount of knowledge. In this

ISSN: 2278-2389
99
era of internet, due to the growth in the collection of images
retrieving of images by user query has become a difficult issue
and therefore, this proposal will facilitates various solutions
regarding different techniques of image retrieving as the basic
ideas have been illustrated by covering all the technicalities.
VIII. CONCLUSION
There have been many advancement in the sector of partial
image retrieval systems especially in the sectors of visual
feature extraction, system designing and multidimensional
indexing. However, there are several issues that indicate
towards its negative points that are related with the practical
use of these systems. Therefore, many researches and
developments are still to go in case of these technological
sectors in the future directions. Single best feature is one of the
prospective works with fully automated image retrieval
systems is one of the key area of concern nowadays. Higher
level concepts of computer vision technologies for the
automatic extraction of object images that are linked with the
lower level features is also an area of research which will
narrow down the semantic gap (Finlayson et al., 1996). Neural
nets, clustering and genetic algorithm are some of the modern
techniques which are needed for the advancement of learning
tools by which off-line as well as on-line processing can be
achieved. Search options in the commercial and academic
sectors that are having a feature of web oriented technologies
in the image retrieval systems are being developed by the
experts. By this, unlimited retrieval of information with a
superior style of organization can be possible. Most of the
web-based search engines are looking forward to this issue
which is an ultimate solution in the sector of text-based
information. The research proposal has been designed to
provide basic knowledge about the joint use of color invariants
and different illumination conditions in the image retrieval
systems as rapid growth has been experienced in the size of
digital image collection.
REFERENCES
[1] G. Finlayson, S. Chatterjee, and B. Funt. Color angular indexing. In
Proceedings of the 4th European Conference on Computer Vision,
Cambridge, England, pages 16{27, 1996.
[2] G. Healey and D. Slater. Computing illumination-invariant descriptors of
spatially filtered color image regions. ieee Transactions on Image
Processing, 6(7):1002{1013, 1997.
[3] J.J. Koenderink and A.J. van Doorn. Representation of local geometry in
the visual system. Biological Cybernetics, 55:367{375, 1987.
[4] T. Lindeberg. Scale-Space Theory in Computer Vision. Kluwer
Academic Publishers, 1994.
[5] R.P.N. Rao and D.H. Ballard. Object indexing using an iconic sparse
distributed memory. In Pro-ceedings of the 5th International Conference
on Computer Vision, Cambridge, Massachusetts, USA,pages 24{31,
1995.
[6] B.M Romeny, L.M.J. Florack, A.H. Salden, and M.A. Viergever. Higher
order diferential structure of images. Image and Vision Computing,
12(6):317{325, 1994.
[7] D. Slater and G. Healey. Combining color and geometric information for
the illumination invariant recognition of 3D objects. In Proceedings of
the 5th International Conference on Computer Vision, Cambridge,
Massachusetts, USA, pages 563{568, 1995.
[8] A.P. Witkin. Scale-space filtering. In Proceedings of the 8th International
Joint Conference on Arti-facial Intelligence, Karlsruhe, Germany, pages
1019{1023, 1983.
[9] X. Wu and B. Bhanu. Gabor wavelets for 3D object recognition. In
Proceedings of the 5th Interna-tional Conference on Computer Vision,
Cambridge, Massachusetts, USA, pages 537{542, 1995.
[10] Kelly, P. M., Cannon, T. M., & Hush, D. R. (1995). Query by image
example: the CANDID approach. SPIE Vol. 2420 Storage and Retrieval
for Image and Video Databases III, 238-248.
[11] Chu, W. W., Cárdenas, A. F., & Taira, R. K. (1995). KMeD: a
knowledge-based multimedia medical distributed database
system. Information Systems, 20(2), 75-96.

Partial Image Retrieval Systems in Luminance and Color Invariants : An Empirical Study

More Related Content

Similar to Partial Image Retrieval Systems in Luminance and Color Invariants : An Empirical Study (20)

More from ijbuiiir1 (20)

Recently uploaded (20)

Partial Image Retrieval Systems in Luminance and Color Invariants : An Empirical Study