As per UGC guidelines an electronic bar code is provided to seure your paper

International Journal for Modern Trends in Science and Technology

Volume 10, Issue 02, pages 329-335.
ISSN: 2455-3778 online
Available online at: http://www.ijmtst.com/vol10issue02.html
DOI: https://doi.org/10.46501/IJMTST1002043

A Parametric Approach Using Active Shape Models for

Facial Landmark Detection

Aswani Lalitha, Irri Vijaya Lakshmi, Dasari Sri Ram, Balla Hemanth, Jammula Ganesh

Department of Electronics and Communications Engineering, Chalapathi Institute of Technology,Guntur, Andhra Pradesh, India

To Cite this Article

Aswani Lalitha, Irri Vijaya Lakshmi, Dasari Sri Ram, Balla Hemanth, Jammula Ganesh, A Parametric Approach
Using Active Shape Models for Facial Landmark Detection , International Journal for Modern Trends in Science and
Technology, 2024, 10(02), pages. 329-335.https://doi.org/10.46501/IJMTST1002043

Article Info
Received: 30 January 2024; Accepted: 21 February 2024; Published: 26 February 2024.

Copyright © Aswani Lalitha et al;. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work
is properly cited.

ABSTRACT
Facial landmark detection is crucial on subsequent tasks related to the face recognition, three-dimensional face
reconstruction, and so on. In this work, we present an active shape model parametric approach for facial landmark detection.
Concretely, we propose to encode the landmark locations by the active shape model (ASM) parameters. Then, we leverage the
power of cascade regression to estimate the ASM parameters. Final landmark locations can be decoded by the ASM parameters.
Through this parametric way of predicting the landmark locations in another domain, it is more compact and effective compared
with existing similar work. Experiments are conducted to validate the effectiveness of our method.

Keywords: active shape model, cascade regression, facial landmark detection

1. INTRODUCTION applied to a wide variety of practical applications

Face is a complex multidimensional structure and including criminal identification, security systems,
needs good computing techniques for recognition. The identity verification etc. Face detection and recognition is
face is our primary and first focus of attention in social used in many places nowadays, in websites hosting
life playing an important role in identity of individual. images and social networking sites. Face recognition and
We can recognize a number of faces learned throughout detection can be achieved using technologies related to
our lifespan and identify that faces at a glance even after computer science.
years. There may be variations in faces due to aging and Features extracted from a face are processed and
distractions like beard, glasses or change of hairstyles. compared with similarly processed faces present in the
Facial features are extracted and implemented through database. If a face is recognized it is known or the system
algorithms which are efficient and some modifications may show a similar face existing in database else it is
are done to improve the existing algorithm models. unknown. In surveillance system if an unknown face
Computers that detect and recognize faces could be appears more than one time then it is stored in database

329 International Journal for Modern Trends in Science and Technology

for further recognition. These steps are very useful in In the recent years, Face recognition get remarkable
criminal identification. In general, face recognition improvement and accuracy to overcome these
techniques can be divided into two groups based on the challenges, but illumination change is still changing. The
face representation they use appearance-based, which objective of this work goal is to implement the system
uses holistic texture features and is applied in face image (model) for a particular face and distinguish it from a
and feature-based, which uses geometric facial features large number of stored faces with some real-time
(mouth, eyes, brows, cheeks etc.), and geometric variations as well. The proposed method is independent
relationships between them The proposed technique is of any judgment of features like lighting problems pose
based on coding and decoding of face images with variations eyes, different facial expressions images, with
emphasis on the significant of local and global features and without Glasses.
of face. 2. LITERATURE REVIEW
Face recognition is one of the major issues in Active shape model (ASM) is an algorithm based on
biometric technology. It identifies and/or verifies a the Point Distribution Model (PDM). In the PDM, the
person by using 2D/3D physical characteristics of the geometric shapes of objects with similar appearances,
face images. Facial landmarks are defined as the such as human faces, human hands, hearts, lungs, etc.,
prominent features that play a discriminative role on the can be represented by the coordinate of several key
facial graphics, such as eye corners, nose tip, and mouth points (landmarks) in series to form a shape vector. The
corners. Facial landmark detection is crucial for a series active shape model treats facial landmarks as a
of tasks related to face, such as facial expression deformable statistical model and then adjusts the
understanding, gaze estimation, and three-dimensional parameters to fit the key points of the given facial image
face reconstruction. The baseline method of face according to the detected local features. There are a lot of
recognition system is the Eigen face by which the goal of feature detection methods that can be used in the ASM,
the eigen face method is to project linearly the image such as Boosted Haar Wavelets [4], Local Binary Patterns
space onto the feature space which has less [5] and mutual information [6]. Regression techniques
dimensionality. One can reconstruct a face image by serve as the alternative to the above methods for local
using only a few eigenvectors which correspond to the search. Unlike the conventional feature detection
largest Eigen values, known as Eigen picture, Eigen face, classifiers in ASM, which learn discrimination function
elastic bunch graph matching and support vector between the local neighborhood and feature appearance,
machine. However, there are still many challenge the regressors used to detect features learn the
problems in face recognition system such as facial relationship between the local neighborhood feature and
expressions, pose variations, occlusion and illumination the displacement of the feature location ground truth.
change. For example, Seise et al. [7] update feature locations with
Those variations dramatically degrade the Relevance Vector Machine regressor. Wimmer et al. [8]
performance of face recognition system. It is evident that regress local Haar Wavelet features to an objective
illumination variation is the most impact of the changes function that can obtain the ground truth of the feature
in appearance of the face images because of its locations. Some studies have compared the performance
fluctuation by increasing or decreasing the intensities of of feature detection methods and regression techniques.
face images due to shadow cast given by different light By comparing Kernel Ridge Regression with a Bayesian
source direction. Therefore, the one of key success is to Classifier approach, Everingham et al. [9] report that the
increase the robustness of face representation against simple classifier method has better performance in eye
these variations. In this proposed method the relevant finding. Cristinacce et.al [10] show that the local feature
information in a face image is feature extracted, encoded regression model performs improved localization and is
and then compared with a face database of models and much more efficient. Some other related works focus on
then classified as known or unknown. improving the shape model for better detection
Face recognition has many challenges due to performance. In the original ASM, the face shape is
illumination variations, large dimensionality, represented through Principal Component Analysis
uncontrolled environments, pose variations and aging. (PCA). Zhou et al. [11] project the facial shapes into a

330 International Journal for Modern Trends in Science and Technology

tangent space and then use Bayesian inference to landmark locations by the ASM parameters. Then, we
estimate both shape and pose parameters. Considering leverage the power of cascade regression to estimate the
that the regressed shape is always the linear combination ASM parameters, which can decode the facial landmark
of all the training shapes, Cao et al. [12] constrain the locations finally. This active shape model parametric
shape model based on the linear combination instead of approach is more compact for the landmark location
PCA. Since that ASM can obtain more accurate detection representation and can partially tackle the problem of
results at a faster speed and is also more robust with overfitting.
regard to illumination, activate shape model has become
3.PARAMETRIC APPROACH USING ACTIVE
a suitable and efficient algorithm for facial landmarks
SHAPE MODELS FOR FACIAL LANDMARK
often implemented on the mobile devices in recent years
DETECTION
[13]. As for the cascade regression-based methods, they
We propose a new framework in the face recognition
learn the regressor at each cascade level to iteratively
System by using Active Shape model (ASM). Initially we
update the landmark positions. The regressor at each
detect the face from the image. After that we extract the
cascade level maps the local features around the current
LBP feature. It is used to find the texture feature for the
key points to the landmark location ground truth. And
face image. The LBP operator assigned a label to every
therefore, the cascade regression-based methods vary in
pixel of a gray level image. The label mapping to a pixel
the input features and regressors. Both SDM [3] and LBF
is affected by the relationship between this pixel and its
[14] use liner regressors at each cascade level. SDM
eight neighbors. Active shape model (ASM) is a
directly uses SIFT features related to the face shape as
statistical model of the shape of objects which iteratively
the input features while LBF learns sparse binarization
deform to fit to an example of the object in a new image.
features in local areas based on the random forest
The shapes are constrained by the PDM (Point
regression model. Discriminative Response Map Fitting
Distribution Model) Statistical Shape Model to vary only
(DRMF) [15] gives a parametric model of the facial
in ways seen in a training set of labeled examples. To
shape, using SVR as the regressor and HOG features as
locate a better position for each point one can look for
the input features. Recently, a cascade regression
strong edges, or a match to a statistical model of what is
framework along with deep learning has achieved
expected at the point. Then weighted matching will be
impressive performance. Deep Convolutional Neural
applied between the input image and database images.
Networks (DCNN) [16] combines coarse-to-fine cascade
and geometric refinement to locate 68 facial landmarks.
Instead of applying deep network directly,
Coarse-to-Fine Auto-encoder Networks (CFAN)
approach [17] cascade a series of consecutive stacked
auto-encoder networks to infer the facial landmarks
from the detected face region nonlinearly. An end-to-end
deep convolutional cascade (DeCaFA) [18] architecture
is introduced to incorporate the landmark-wise attention
maps and intermediate supervisions into the deep Figure 1: Proposed Model for Facial Landmark
cascade convolution network for landmark detection. To Detection
tackle the problem of landmark detection under MODULES in the processing are given by
occlusion, Wan et al. [19] propose to integrate a deep  Preprocessing
regression module and a deocclusion module into the  Normalization
cascade regression framework. However, the  Active shape model
performance of deep learning-based methods highly  Feature Extraction
relies on the scale of training samples. In addition, these  Recognition
methods are more likely to overfit the data. In this work, 3.1 Module Description
an active shape model parametric approach is proposed Preprocessing
for facial landmark detection. We first encode the

331 International Journal for Modern Trends in Science and Technology

 In noise removal process, initially we convert the However, the LBP operator is not directly affected by the
image in gray. And then we filter the noise from the gray value of Z0, so we can redefine the function as
image. In Filtering we are applying Gaussian filtering to following:
our input image. Gaussian filtering is often used to T = t (Z0-Z1, Z0-Z2, …, Z0-Z8).
remove the noise from the image. Here we used wiener2 To simplify the function and ignore the scaling of grey
function to our input image. Gaussian filter is level, we use only the sign of each element instead of the
windowed filter of linear class, by its nature is weighted exact value. So the operator function will become:
mean. Gaussian filter is named after famous scientist T = t (s(Z0-Z1), s(Z0-Z2), …, s(Z0-Z8)).
Carl Gauss because weights in the filter calculated Where the s(.) is a binary function, defined as
according to Gaussian distribution. s(x) =1, x>=0; S(x) = 0, otherwise.
Normalization Histogram features
Normalization is a process that changes the range of An image histogram is a type of histogram that acts as
pixel intensity values. Illumination changes caused by a graphical representation of tonal distribution describes
light sources at arbitrary positions and intensities the distribution of various bright and dark tones with in
contribute to a significant amount of variability. To an image. During the scanning or image editing stage
address this issue, we present a new method for tones can be redistributed lightening a dark image (or)
performing image normalization. The method used to darkening a bright image. This histogram plots the no. of
remove shadows and specularities from images. All the pixels for each tonal value. by looking at the histogram
shadowed regions are grayed out to a uniform color, for a specific image a person will be able to judge the
eliminating soft shadows and specularities and hence entire tonal distribution.
creating an illumination invariant signature of the Image histograms are present on many modern digital
original image. cameras. The horizontal axis of the graph represents the
Active Shape model tonal variations and the vertical axis represents the no.of
Active shape models (ASMs) are statistical models of pixels in that particular tone. For this histogram we are
the shape of objects which iteratively deform to fit to an assuming a discrete function h(rk)=nk Here rk is the kth
example of the object in a new image. The shapes are gray level and nk is the no. of pixels in the image at the
constrained by the PDM (point distribution model) gray level rk.
Statistical shape model to vary only in ways seen in a
training set of labelled examples. The shape of an object
is represented by a set of points (controlled by the shape
model). The ASM algorithm aims to match the model to
a new image. It works by alternating the following
steps:Look in the image around each point for a better
position for that point. Update the model parameters to
best match to these new found positions.
Feature Extraction
Initially we separate the image as patches. For each
patch of image, we apply the LBP (Local Binary Pattern). Figure 2: Different Types of Histogram Images with
The LBP operator assigned a label to every pixel of a Different Contrasts
gray level image. The label mapping to a pixel is affected Recognition
by the relationship between this pixel and its eight Here the recognition process is identified by the
neighbors of the pixel. If we set the gray level image is I, weighted matching. The Euclidean distance for LBP
and Z0 is one pixel in this image. So we can define the based histogram features is computed for the test feature
operator as a function of Z0 and its neighbors, Z1, …, Z8. with the database features. The similarity is identified
And it can be written as: between the features. Finally identified image is
T = t (Z0, Z0-Z1, Z0-Z2, …, Z0-Z8). displayed.
4. RESULTS& DISCUSSION

332 International Journal for Modern Trends in Science and Technology

Data Set
We have employed a set of images These images have
been logically chosen by art historians in order to
address different tasks such as (a) to test the relation of
an unmediated image of the subject, e.g., a death mask to
a work of portrait art like a painting, (b) to analyze a
number of portraits of different sitters by the same artist
to model artist’s style, (c) to verify if the identity of the
ambiguous subject in a given image is same as that of a
known subject in a reference image.
Identity Verification Filtering
In order to examine the validity of the chosen
approach, we consider similarity scores of the test image
with artworks known to depict persons different from
the one depicted in reference image. We call these
images as distracters. In cases where enough works of
the same artist is not available, we consider similar
works of other artists. If a test image indeed represents
the same sitter as in the reference image, not only should
its score with the reference image be modelled by the
match distribution, but also its scores with distracter
faces should be modelled by the non-match distribution.
The results of the proposed system are shown below.
Selection of input image Normalization

Active Shape Model

Recognition
Input image

333 International Journal for Modern Trends in Science and Technology

 International Conference on Automatic Face and Gesture
Recognition (FGR06). IEEE, 2006, pp. 441–446.
[7] D. Cristinacce and T. F. Cootes, “Boosted regression active shape
models.” in BMVC, vol. 2. Citeseer, 2007, pp. 880–889.
[8] Y. Zhou and H.-j. Zhang, “Bayesian tangent shape model:
Estimating shape and pose via bayesian inference,” in IEEE
Conf. on CVPR. Citeseer, 2003.
[9] X. Cao, Y. Wei, F. Wen, and J. Sun, “Face alignment by explicit
shape regression,” International Journal of Computer Vision,
vol. 107, no. 2, pp. 177–190, 2014.
[10] Y.-H. Lee, C. G. Kim, Y. Kim, and T. K. Whangbo, “Facial
landmarks detection using improved active shape model on
android platform,” Multimedia Tools and Applications, vol. 74,
no. 20, pp. 8821–8830, 2015.
Advantages: [11] S. Ren, X. Cao, Y. Wei, and J. Sun, “Face alignment at 3000 fps via
1. Security levels will be significantly improved. regressing local binary features,” in Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition, 2014,
2. The integration process is easy and flawless
pp. 1685–1692.
3. High accuracy allows avoiding false identification [12] A. Asthana, S. Zafeiriou, S. Cheng, and M. Pantic, “Robust
4. Facial Recognition System is fully automated discriminative response map fitting with constrained local
5. Time fraud will be excluded models,” in Proceedings of the IEEE conference on computer
vision and pattern recognition, 2013, pp. 3444–3451.
Applications
[13] E. Zhou, H. Fan, Z. Cao, Y. Jiang, and Q. Yin, “Extensive facial
1.Historical Persons Database landmark localization with coarse-to-fine convolutional network
2.Politics cascade,” in Proceedings of the IEEE international conference on
3.Education computer vision workshops, 2013, pp. 386–391.
[14] J. Zhang, S. Shan, M. Kan, and X. Chen, “Coarse-to-fine
4.Industrial
auto-encoder networks (cfan) for real-time face alignment,” in
5.Security European conference on computer vision. Springer, 2014, pp.
6.Military 1–16.
[15] A. Dapogny, K. Bailly, and M. Cord, “Decafa: deep
Conflict of interest statement convolutional cascade for face alignment in the wild,” in
Proceedings of the IEEE International Conference on Computer
Authors declare that they do not have any conflict of
Vision, 2019, pp. 6893–6901.
interest. [16] J. Wan, J. Li, Z. Lai, B. Du, and L. Zhang, “Robust face alignment
by cascaded regression and de-occlusion,” Neural Networks,
REFERENCES vol. 123, pp. 261–272, 2020.
[1] P. Viola and M. Jones, “Rapid object detection using a boosted [17] Ravikiran, D. N., & Dethe, C. G. (2018). Improvements in Routing
cascade of simple features,” in Proceedings of the 2001 IEEE Algorithms to Enhance Lifetime of Wireless Sensor
computer society conference on computer vision and pattern Networks. International Journal of Computer Networks &
recognition. CVPR 2001, vol. 1. IEEE, 2001, pp. I–I. Communications (IJCNC), 10(2), 23-32.
[2] T. Ahonen, A. Hadid, and M. Pietikainen, “Face recognition with [18] Ravikiran, D. N., & Dethe, C. G. Fuzzy Rule Selection using
local ¨ binary patterns,” in European conference on computer LEACH Algorithm to Enhance Life Time in Wireless Sensor
vision. Springer, 2004, pp. 469–481. Networks. Advances in Wireless and Mobile Communications.
[3] N. D. Dowson and R. Bowden, “Simultaneous modeling and ISSN, 0973-6972.
tracking (smat) of feature sets,” in 2005 IEEE Computer Society [19] Rajesh, G., Thommandru, R., & Subhani, S. M. DESIGN AND
Conference on Computer Vision and Pattern Recognition IMPLEMENTATION OF 16-BIT HIGH SPEED CARRY SELECT
(CVPR’05), vol. 2. IEEE, 2005, pp. 99–105. PARALLEL PREFIX ADDER.
[4] M. Seise, S. J. McKenna, I. W. Ricketts, and C. A. Wigderowitz, [20] Polanki, K., Purimetla, N. R., Roja, D., Thommandru, R., &
“Learning active shape models for bifurcating contours,” IEEE Javvadi, S. Predictions of Tesla Stock Price based on Machine
transactions on medical imaging, vol. 26, no. 5, pp. 666–677, Learning Model.
2007. [21] Thommandru, R. A PROSPECTIVE FORECAST OF BRAIN
[5] M. Wimmer, F. Stulp, S. Tschechne, and B. Radig, “Learning STROKE USING MACHINE LEARNING TECHNIQUES.
robust objective functions for model fitting in image [22] Rajesh, G., Raja, A., & Thommandru, R. OPTIMIZATION OF
understanding applications.” in BMVC, vol. 3, 2006, pp. MINIATURIZED MICROSTRIP PATCH ANTENNAS WITH GA.
1159–1168. [23] Vellela, S. S., & Balamanigandan, R. (2022, December). Design of
[6] M. Everingham and A. Zisserman, “Regression and classification Hybrid Authentication Protocol for High Secure Applications in
approaches to eye localization in face images,” in 7th Cloud Environments. In 2022 International Conference on

334 International Journal for Modern Trends in Science and Technology

 Automation, Computing and Renewable Systems (ICACRS) (pp.
408-414). IEEE.
[24] Vellela, S. S., & Balamanigandan, R. (2024). Optimized clustering
routing framework to maintain the optimal energy status in the
wsn mobile cloud environment. Multimedia Tools and
Applications, 83(3), 7919-7938.
[25] Praveen, S. P., Sarala, P., Kumar, T. K. M., Manuri, S. G., Srinivas,
V. S., & Swapna, D. (2022, November). An Adaptive Load
Balancing Technique for Multi SDN Controllers. In 2022
International Conference on Augmented Intelligence and
Sustainable Systems (ICAISS) (pp. 1403-1409). IEEE.
[26] Priya, S. S., Vellela, S. S., Reddy, V., Javvadi, S., Sk, K. B., & Roja,
D. (2023, June). Design And Implementation of An Integrated IOT
Blockchain Framework for Drone Communication. In 2023 3rd
International Conference on Intelligent Technologies
(CONIT) (pp. 1-5). IEEE.
[27] Vellela, S. S., & Balamanigandan, R. An intelligent sleep-awake
energy management system for wireless sensor network.
Peer-to-Peer Netw. Appl.(2023).
[28] Addepalli, T., Babu, K. J., Beno, A., Potti, B. M. K., Sundari, D. T.,
& Devana, V. K. R. (2022). Characteristic mode analysis of two
port semi‐circular arc‐shaped multiple‐input‐multiple‐output
antenna with high isolation for 5G sub‐6 GHz and wireless local
area network applications. International Journal of
Communication Systems, 35(14), e5257.
[29] Srija, V., & Krishna, P. B. M. (2015). Implementation of
agricultural automation system using web & gsm
technologies. International Journal of Research in Engineering
and Technology, 04 (09), 385-389.
[30] Potti, D. B., MV, D. S., & Kodati, D. S. P. (2015). Hybrid genetic
optimization to mitigate starvation in wireless mesh
networks. Hybrid Genetic Optimization to Mitigate Starvation in
Wireless Mesh Networks, Indian Journal of Science and
Technology, 8(23).
[31] Potti, B., Subramanyam, M. V., & Prasad, K. S. (2013). A packet
priority approach to mitigate starvation in wireless mesh network
with multimedia traffic. International Journal of Computer
Applications, 62(14).
[32] Potti, B., Subramanyam, M. V., & Satya Prasad, K. (2016).
Adopting Multi-radio Channel Approach in TCP Congestion
Control Mechanisms to Mitigate Starvation in Wireless Mesh
Networks. In Information Science and Applications (ICISA)
2016 (pp. 85-95). Springer Singapore.

335 International Journal for Modern Trends in Science and Technology