INTRODUCTION on texts as well. So it remains a

INTRODUCTION

 

The advancement
of Technology in the 21st Century makes a common intent of every
action to be based on real time analysis and impeccable accuracy. In case of
Image retrieval or video-text detection and recognition, it poses a matter of
even deeper insight while dealing with situations concerning security and
surveillance like directing Blind people on road or retrieving the
alpha-numeric from number plate of a Car. 
The Quality of the videos are significantly degraded owing to factors
like motion blur, non- uniform illumination, complex background and text
movement. Other than that the variance of lightings and perspective distortions
has a major negative impact on texts as well. So it remains a question of an
attainable accuracy among the researchers. Text being one of the prime channel
of communication, the retrieval of textual information from scene images and
videos frames has gained the attention of the researchers from the very
beginning. A robust text detection framework is required to detect text information
from a scene images and video frames. Many theories and applications proposed,
the accuracy of text detection from a video frame still remains a challenge as
the inputs are unrestricted to colors, fonts and even orientation.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

The methodology
of text detection can be resolved into namely three advances: connected
component–bases, edge based and texture based. The first method uses the
principle of color quantization and splitting to connect or group the adjacent
pixel of similar color into connected components. But the factors like color
bleeding and the low contrast of text lines defy in manufacturing the complete
image and hence is not applicable for video images. In order to overcome this
impending problem, edge based techniques are proposed. In this methodology both
the horizontal as well as the vertical profile of the edge map are analyzed. Though
expedited, this process of text detection in video frames has the probability
of producing false positive at large in instances of complex background. In
order fix such problems, texture based approach is applied which considers the
text regions as texture or couture. Feature extractions in this texture based approach
is done using Fast Fourier Transform, discrete cosine transform, wavelet
decomposition, and Gabor filters. And such method involves classifiers like SVM
and neural networks.  But these
classifiers needs extensive training data of text and non-text for attaining
higher accuracy.

                Detection
of text with high precision on multi-oriented dimensions without restriction on
the background, alignment, and contrast and re-call still remains a difficult
task. Existing models focused on horizontal text–orientation fails when applied
on multi-oriented text frames. Successful research on this field is much
limited due to above mentioned restrictions. Accordingly, in this paper, we
have deduced a methodical model which will handle linear text in multiple orientation
as well as curved texts. In addition to which we have proposed a HMM based
verification to attain higher accuracy.

                The
proposed text detection framework which involves both linear text and curve
texts in any frames is developed from the very base to an advanced infrastructure.
For better filtering, Laplacian of Gaussian filter is used in the proposed
method. One of the novelties of this method belongs to the domain of skeletal featuring
of the texts which may reside in nonlinear dimensions and HMM based
verification of the text resulting in attaining a optimum level accuracy.  

 

x

Hi!
I'm Brent!

Would you like to get a custom essay? How about receiving a customized one?

Check it out