International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 10 | Oct 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 40
Comparative Study of Different Techniques for Text as well as Object
Detection from Real Time Images
Monika Kapoor
1
, Er. Saurabh Sharma
2
1
Student, Dept of computer science Engineering, Sri Sai University Palampur (H.P) India
2
Assitant Professor, Dept. of Computer science and Engineering, Sri Sai University Palampur (H.P) India
---------------------------------------------------------------------------***---------------------------------------------------------------------------
Abstract - Since the introduction of artificial intelligence, it
has been a major curiosity of the developers to make
computers think more like humans. Artificial Neural
Networks was developed with the view in mind to make
computers to do so. These intelligent machines can be
used in the field of robotics, medicines, industry. Since the
introduction of image recognition algorithms by face-book
AI researchers image identification and recognition has
become a curios field among the developers. The main
objective of the image identification software(s) is to
differentiate the components of the image or the scene and
tell them apart. Many algorithms such as OCR, RCNN, Mask
RCNN, Fast RCNN, Faster RCNN algorithms were
developed to identify and classify images into the
categories as desired by the programmer. Here in this
paper, we would throw some light on the aforementioned
different types of Image Processing algorithms and try to
determine which of these are best suited. Researchers
have proved that Mask RCNN, which is a better version of
Faster RCNN proved to the best suited algorithm in the
field of object detection in the real time.
Keywords: Artificial Neural Networks, Machine Learning,
Object Identification, Scene Identification, OCR, Mask
RCNN, RCNN.
1. INTRODUCTION
Realworld applications are looking forward to make use of
computers to make tasks easier for people in real world
purposes. Due to its easiness in the usage computers are
used in wide variety of fields such as robotics, medicine
and advanced computing. Likewise, if robots are feed data
about their surroundings then they can be better informed
about their nearby environments to handle the situations.
In a similarly way, if a computer can been taught how to
differentiate and recognize tumor cells then it can be used
for the identification of the tumors and cancer cells from
the hundreds of pictures and can reduce the labor of
manually identifying the tumors[1]. Likewise, in the case of
the self-driven cars it becomes important for the system to
identify the objects not only in the still images but in real
time and take measures accordingly. Object detection aims
to learn the concept of visual models in an image[2]. The
ability to model the various deformations, inclusions, and
other class variations while handling the large amount of
data simultaneously under several conditions, is the main
objective through this type of study.
Machine Learning (ML) [3] is a branch of artificial
intelligence that is targeted towards the training of
machines; designing the algorithms to handle robots etc.
these machines learn to operate on their own after training
on datasets. Machines are taught to take data driven
decisions through self-calculation and self-observation
instead of being exclusively programmed for a special task
they are programmed for performing a number of tasks
simultaneously. The a self-iterating algorithm is developed
in such a way that it learns and improves itself. If a new
input data is encountered by the ML algorithm, it makes a
forecast based on the model. This forecast is validated for
precession and if the precision is within acceptable choice,
the Machine Learning algorithm is deployed. If the
precision value is not accepted, the Machine Learning
algorithm is trained to perform a number of iterations
couple of times with an enlarged training data set.
Figure 1: Machine learning algorithm [4]
The question arises as how to train the machines so that
they can be made to think like humans. The answer to this
question is provided by artificial neural networks (ANNs).