Biometric Authentication


Biometrics deals with recognizing people based on their physiological or behavioral characteristics. Our work primarily concentrates on three different aspects in biometrics:

  1. Enhancing Weak Biometrics for Authentication: Weak biometrics (hand-geometry, face, voice, keystrokes) are the traits that possess low discriminating content and they change over time for each individual. However, there are several traits of weak biometrics such as social acceptability, ease of sensing, and lack of privacy concerns that make weak biometrics ideally suited for civilian applications. Methods that we developed can effectively handle the problems of low discriminative power and low feature stability of weak biometrics, as well as time-varying population in civilian applications.
  2. Writer Identification from Handwritten Documents: Handwriting is a behavioural biometric that contains distinctive traits aquired by a person over time. Traditional approaches to writer identification tries to compute feature vectors that capture traits of handwriting that are known to experts as discriminative. In contrast we concentrate on automatic extraction of features that are suitable to specific applications such as writer identification in civilian domain and in problems such as forgery and repudiation in forensics.
  3. Use of Camera as a Biometric Sensor: Camera has been used for capturing face images for authentication in the past. However, with biometrics traits such as fingerprints and iris, a specialized sensor is often preferred due to the high quality of data that they provide. Recent advances in image sensors have made digital cameras both inexpensive and technically capable for achieving high quality images. However, many problems such as variations in pose, illumination and scale restrict the use of cameras as sensors for many biometric traits. We are working on the use of models of imaging process to overcome these problems, to capture high quality data for authentication.

Enhancing Weak Biometric based Authentication


Weak biometrics (hand-geometry, face, voice, keystrokes) are the traits which possess low discriminating content and they change over time for each individual. Thus they show low accuracy of the system as compared to the strong biometrics (eg. fingerprints, iris, retina, etc.) However, due to exponentially decreasing costs of the hardware and computations, biometrics has found immense use in civilian applications (Time and Attendance Monitoring, Physical Access to Building, Human-Computer Interface, etc.) other than forensics (e.g. criminal and terrorist identification). Various factors need to be considered while selecting a biometric trait for civilian application; most important of which are related to user psychology and acceptability, affordability, etc. Due to these reasons, weak biometric traits are often better suited for civilian applications than the strong biometric traits. In this project, we address issues such as low and unstable discriminating information, which are present in weak biometrics and variations in user population in civilian applications.

schdaDue to the low discriminating content of the weak biometric traits, they show poor performance during verification. We have developed a novel feature selection technique called Single Class Hierarchical Discriminant Analysis (SCHDA), specifically for authentication purpose in biometric systems. SCHDA builds an optimal user-specific discriminant space for each individual where the samples of the claimed identity are well-separated from the samples of all the other users.

The second problem which leads to low accuracy of authentication is the poor stability or permanence of weak biometric traits due to various reasons (eg. ageing, the person gaining or losing weight, etc.) Civilian applications usually operate in cooperative or monitored mode wherein the users can give feedback to the system on occurrence of any errors. An intelligent adaptive framework is used, which uses feedback to incrementally update the parameters of the feature selection and verification framework for each individual.

The third factor that has been explored to improve the performance of an authentication system for civilian applications is the pattern of participation of each enrolled user. As the new users are enrolled into the system, a degradation is observed in performance due to increasing number of users. An interesting observation is that although the number of users enrolled into the system is very high, the number of users who regularly participate in the authentication process is comparatively low. We model the variation in participating population using Markov models. The prior probability of participation of each individual is computed and incorporated into the feature selection framework, providing more relevance to the parameters of regularly participating users. Both the structured and unstructured modes of variation of participation are explored.

Text Independent Writer Identification from Online Handwriting

Handwriting Individuality is a quantitative measure of writer specific information that can be used to identify authorship of the documents and study of comparison of writing habits, evaluation of the significance of their similarities and differences. It is an discrimitive process like fingerprint identification, firearms identification and DNA analysis. Individuality in handwriting lies in the habits that are developed and become consistant to some degree in the process of writing.

Discriminating elements of handwriting lies in various factors such as i) Arrangement, Connections, Constructions, Design, Dimensions, Slant or Slope, Spacings, CLass and choice of allographs, 2) Language styles such as Abbreviation, Commencements and terminations, diacritics and punctuation, line continuity, line quality or fluency, 3) Physical traits such as pen control, pen hold, pen position, pen pressure and writing movement, 4) Consistancy or natural variations and persistance, and 4) Lateral expansion and word proportions.

The framework that we utilize tries to capture the consistent information at various levels and automatically extract discriminative features from them.

Features of our Approach:clusters

  • Text-independent algorithm: Writer can be identified from any text given in underlined script. Comparison of features are not done for the similar charcters.
  • Script dependent framework: Applicablity is verified on different scripts like Devanagiri, Arabic,Roman, Chinese and Hebrew.
  • Use of Online Information: Online data is used for verification purpose. Offline information is also applicable with similar framework with appropriate change in feature extraction.
  • Authentication with small amount of data: Around 12 words in Devanagiri we get accuracy of 87%.

Underlying process of identification:


  • Primitive Definition:

    Primitives are the discrimitive features of handwriting documents. First step is to identify primitive. Primitives can be individuality features like size, shape, distribution of curves in handwritten document. We choose subcharcter level curves as basic primitives

  • Extraction and Representation of primitive:

    Extraction of primitive is done using velocity profile of the stroke shown in the figure. Minimum velocity points are critical points of primitive. Primitives are extracted using size and shape features as shown in diagram.

  • Identification of Consistant Primitives:

    Repeating curves are consitent primitives. To extract consistent curves, unsupervised clustering algorithm is used to cluster them into different groups.

  • Classification:

    Variation in distribution, size and shape of curves in each cluster is used to discriminate writer from other writers.

Related Publications

  • Vandana Roy, C. V. Jawahar: Feature Selection for Hand-Geometry based Person Authentication, in Proceedings of International Conference on Advanced Computing and Communication, Coimbatore, India, Dec. 2005.
  • Vandana Roy, C. V. Jawahar: Hand-Geometry Based Person Authentication Using Incremental Biased Discriminant Analysis, in Proceedings of National Conference on Communications, New Delhi, Jan. 2006.
  • Anoop Namboodiri, Sachin Gupta: Text Independent Writer Identification for online Handwriting, in Proceedings of the International Workshop on Frontiers in Handwriting Recognition (IWFHR'06), La Baule, France, October 2006.
  • Vandana Roy, C. V. Jawahar: Modeling Time-Varying Population for Biometrics, To appear in Proceedings of International Conference on Computing: Theory and Applications, Kolkata, India, March 2007.

Associated People


A Rule-based Approach to Image Retrieval



Published in:      IEEE Region 10 Conference on Convergent Technologies, TENCON 2003

Authors:              Dhaval Mehta,    E.S.V.N.L.S.Diwakar,,    C. V. Jawahar



Virtual Textual Representation for Efficient Image Retrieval



Published in:      3rd International Conference on Visual Information Engineering, VIE 2006

Authors:              P Suman Karthik,    C. V. Jawahar



The state of the art in contemporary visual object categorization and classification is dominated by “Bag Of Words” approaches. These use either discriminative or generative learning models to learn the object or scene model. In this paper, we propose a novel “Bag of words” approach for content based image retrieval. Images are converted to virtual text documents and a new relevance feedback algorithm is applied on these documents. We explain how our approach is fundamentally different to existing ones and why it is ideally suited for CBIR. We also propose a new hybrid relevance feedback learning model. This merges the best of generative and discriminative approaches to achieve a robust and discriminative visual words based description of a visual concept. Our learning model and “Bag Of Words” approach achieve a balance between good classification and efficient image retrieval.


 paper vie2 



Effecient Region Based Indexing and Retrieval for Images with Elastic Bucket Tries


Published in:      International Conference on Pattern Recognition, ICPR 2006

Authors:              P Suman Karthik,    C. V. Jawahar



Retrieval and indexing in multimedia databases has been an active topic both in the Information Retrieval and com- puter vision communities for a long time. In this paper we propose a novel region based indexing and retrieval scheme for images. First we present our virtual textual description using which, images are converted to text documents con- taining keywords. Then we look at how these documents can be indexed and retrieved using modified elastic bucket tries and show that our approach is one order better than stan- dard spatial indexing approaches. We also show various operations required for dealing with complex features like relevance feedback. Finally we analyze the method compar- atively and and validate our approach.

paper ebt 

Private Content Based Image Retrieval


For content level access, very often database needs the query as a sample image. However, the image may contain private information and hence the user does not wish to reveal the image to the database. Private Content Based Image Retrieval (PCBIR) deals with retrieving similar images from an image database without revealing the content of the query image . not even to the database server.

PCBIR Overview

  1. PCBIR

    The user extracts the feature vector of the query image,say fquery.
  2. The user first asks the database to send the information at the root node.
  3. Using fquery and the information received, the user decides whether to access the left subtree or the right subtree.
  4. In order to get the data at the node to be accessed, the user frames a query Qi where i indicates the level in which the node occurs.(Please note that the root is at level 0)
  5. The database returns a reply Ai for the query Qi.
  6. The user performs a simple function f(Ai) to obtain the information at the node. If the node is a leaf node, user adds the information to the results else goto step 3.

PCBIR on a Binary Search Tree

Preprocessing Step

  • Consider a natural number N = p. q where p, q are large prime numbers.
  • Construct a set Zn*={x| 1 < x < N, gcd(N,x)=1}
  • `y` is called a Quadratic Residue (QR), if x | y = x2 and else `y` is called a Quadratic Non-Residue (QNR).
  • Construct a set YN with equal number of QRs and QNRs
  • BST
  1. Suppose the user wants to extract the kth node at ith level in the tree. The ith level shall contain 2i nodes.
  2. All the nodes are treated as a 2D array of mxn dimensions where m x n = 2i.
  3. If the user wants to access the kth node, now he has to access the node at (k/n, k mod n) (say this is (x,y)).
  4. To get the data at node (x,y), the user frames a query Qi of length m, with a QNR at position x and rest of the values being QRs.
  5. The database computes a reply Ai for the query Qi by forming a matrix, in which if the value at the node is 1, its replaced by square of the value in query else its replaced by the value. Then the matrix is multiplied column wise to obtain the reply Ai of length n.
  6. The user then checks whether the value at the yth position is a QR or a QNR. If it is a QR the value is 1 else it is 0. This is due to the properties below
    • QNR x QNR = QR
    • QNR x QR   = QNR
    • QR   x QR   = QR
  7. The above protocol is run for every level. Since the database is oblivious about the node which the user is trying to extract, the query path is hidden securing the privacy of the query image.

The algorithm is based on the Quadratic Residuosity Assumption which states that:

Given a number `y` belongs to YN, it is predictably hard to decide whether `y` is a QR or a QNR..

Extension to other Hierarchical Structures

Hierarchical structures primarily vary in

  • Number of nodes at each level.
  • Information at a node.

If we can take care of these two, the algorithm can be extended to any hierarchical structure used for indexing. Our algorithm requires a m x n matrix to be formed and a set of nodes at a level can be easily converted in to such a format (irrespective of the number of the nodes). Moreover the data transfer takes place in binary format and any data in the intermediate nodes of an indexing structure can be represented in binary format. So if the user has the data about the indexing structure and the format of the information stored at a node, the algorithm can be simulated for any hierarchical structure.

Experiments and Results

Experiments were conducted on popular indexing schemesto check the feasibility and applicability of the PCBIR algorithms. The experiments were conducted using a GNU C compiler on a 2 GHz Intel Pentium M process with 4 GigaBytes of main memory. Information regarding the indexing structures and the datasets.

  • KD-Tree and Corel Database

The Corel dataset consisted of 9907 images, scaled to 256 x 128 resolution. The color feature was used to index the images. The feature vector (i.e color histogram) was 768 in length (256 values each for RGB). A KD tree usually stores the ssplit value and split dimension at non leaf nodes. Each value was represented in 32 bits, thus the total information at each intermediate/non-leaf node was 64 bits. Sample results have been shown below. The image with a black box is the query image and the other images are the top 4 retrieved images. The Corel dataset consisted of 9907 images, scaled to 256 x 128 resolution. The color feature was used to index the images. The feature vector (i.e color histogram) was 768 in length (256 values each for RGB). A KD tree usually stores the ssplit value and split dimension at non leaf nodes. Each value was represented in 32 bits, thus the total information at each intermediate/non-leaf node was 64 bits. Sample results have been shown below. The image with a black box is the query image and the other images are the top 4 retrieved images.

corel result1

corel result2

The average retrieval time for a query was 0.596 secs. The time was obtained by amortization over 100 queries.

  • Vocabulary Tree and Nister Dataset

SIFT featues in an image were used to obtain a vocabulary of visual words. The vocabulary was of 10,000 visual words. The Nister dataset consisting of 10,200 images was used. The dataset consists of 2540 object photographed from 4 different angles and lighting conditions. The vocabulary tree was constructed using a training set of 2000 images which were picked from the dataset itself. Sample results have been shown below. The image with a black box is the query image and the other images are the top 4 retrieved images.




The average retrieval time for a query was 0.320 secs. The time was obtained by averaging over 50 queries which were amortized.Since the vocabulary tree allows us to change the size of the vocabulary, retrieval times under various vocabulary sizes were recorded in the following graph.

  •  LSH and Corel Dataset

LSH consisted of a set of 90 hash functions each with 450 bins on average. The images in each bin were further subdivided by applying LSH again - Thus gaining an hierarchy of 2 levels. The average retrieval time was 0.221.The algorithm can also be used to achieve partial privacy which is indicated by 'n', also known as the confusion metric. When its value is 1, the retrieval is completely private and 0 indicates that its non-private image retrieval.

 cm  graph


  • Scalability

Due to the lack of very large standard image datasets, we had to test the scalability of our system using synthetic datasets. The retrieval times for various dataset sizes is provided in the table below.


Dataset Size Average Retrieval Time
2^10 0.005832
2^12 0.008856
2^14 0.012004
2^16 0.037602
2^18 0.129509
2^20 0.261255


Related Publications

Shashank Jagarlamudi, Kowshik Palivela, Kannan Srinathan, C. V. Jawahar : Private Content Based Image Retrieval. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008.

Associated People 

  • Shashank Jagarlamudi
  • Kowshik PalivelaK
  • Kannan Srinathan
  • C.V. Jawahar




Poster [ppt]



Presentation [ppt]