Text Extraction From Digital Images With Text To Speech Conversion and Language Translation
Text Extraction From Digital Images With Text To Speech Conversion and Language Translation
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Text Extraction from Digital Images with Text to
Speech Conversion and Language Translation
Akhil Chandran Miniyadan Nithya G P
Assistant Professor, Department of Information Technology Assistant Professor, Department of Information Technology
College of Engineering Trikaripur College of Engineering Thalassery
Kasaragod, Kerala, India Kannur, Kerala, India
Abstract:- In this age of digital information, the need for In this paper, we propose a system which extracts text
digitalizing anything and everything is a growing need. from a given picture and then convert it into speech and it
Today most of the information is available either on paper can also be converted into many other languages. The system
or in the form of photographs. Tracking and modifying tries to combine the needed text manipulations in one single
information from images is inconvenient and time- product.
consuming. Thus we need to extract textual images into
editable text. We do have technologies to extract text but II. EXISTING METHODOLOGY
they are mainly against clean backgrounds and it seemed
to generate erroneous results. Thus, there is a need for a Now, the old method to perform the text to speech con-
system to extract text from general backgrounds with version requires a web camera to acquiring an image and
more accuracy. Generating such text can be utilized as an converting it into a text document using Optical Character
input to a TTS Text-to-Speech where it converts any text Recognition (OCR). The next stage involves natural language
into a speech signal and later on translated into the processing and digital signal processing for converting the text
desired language. Our market is beamed with text into speech using Text to Speech synthesizer (TTS). The
extraction, TTS and translation products but is often in following paragraphs discuss various methods in the field.
separate products. Our aim to achieve is combining all
these text manipulations in one product and to be able to The method proposed in [1] is a combination of scene
sync with existing text modifying products. text detection and scene text recognition algorithm. An image
with text is given as input, preprocessing methods are used to
Keywords:- Optical Character Recognition, Firebase ML Kit, remove noises. Binarization helps in identifying text from an
Binarization, TTS, Discrete Contour Evolution. image. Thinning and scaling is performed by connectivity
algorithm if any data is lost during preprocessing. The
I. INTRODUCTION approach in [1] uses a character descriptor to separate text
from an image. The detected text is converted using a
Information today has been highly graphical and are descriptor and wavelet feature. Sibling of each character is
stored in the form of images or in videos. Yet recent calculated using an adjacent character grouping algorithm.
technology is restricted to how to retrieve those Stroke-related features are extracted using skeletons and
informational texts from the image. That’s why text character boundaries. There are 3 main steps in the
extraction plays a vital role in many applications that implementation as follows: Given a synthesized patch from a
include information retrieval, digital library, multimedia training set, we obtain character boundary and character
systems, and many more. Text can be of immense use if can skeleton by applying discrete contour evolution (DCE) and
be converted to audio, especially for reducing visual skeleton pruning on the basis of DCE.
reliability. Text-to-speech (TTS) is a process of producing
spoken word from the text. As the world has grown into a In robust algorithm [2], proposes a new text detection
global village, the diversity in native languages shouldn’t and extraction method that overcomes the weakness of
make anyone from experience life as an outsider. Thus previous approaches. The input image is first transformed into
language conversion to the desired language of the user can a binary image and edge detection is applied. Instead of
be of great use, in another lingual atmosphere. performing a simple thresholding method, maximally stable
external regions (MSER) are detected. These regions contain
Text manipulations have been always in trend and in the text components and are appointed as white pixels.
need. Text to speech systems was initially developed to assist However, the resulting binary image does not reveal the exact
the visually impaired by offering a computer-generated boundaries of the text. For this reason, MSER binary image is
spoken voice that would read out loud the text. It beneficial, enhanced by performing a thresholding operation on each
especially for blind people as they will be able to understand connected component. Edges are then detected and fed into a
what is written by hearing it. Language translation is very stroke width detector where strokes, stroke widths, and
helpful, especially to understand signboards. connected components are found and filtered. In this using a
robust algorithm, that proved to be effective on blurred images
and noisy images as well.
IJISRT21MAR439 www.ijisrt.com 565
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Image processing done by implementing Tess-Two 2) Image Processing: The captured image, in this stage, is
tesseract library [3] in Android studio and Google Vision API. processed. For this, we implement efficient text recognizing
Different Image processing techniques will be used to enhance functions in firebase ML Kit’s text recognition API. The
the image quality by removing noise from an image, process is efficient and tends to retrieve standard and uniform
improving contrast and brightness of an image and a cropping texts from the image. The textual image is scanned to identify
image feature will also be provided for increasing the accuracy elements of recognition like block, symbol, word and
of detection of text from an image. For translating to Braille paragraph by the region bounded by coordinates of the area.
python library is used. After recognizing the elements of text, they are compared with
supported languages available in Google Cloud platforms.
In [4] a text which is compatible with an android phone This extracted text can be the input of the next module of TTS.
is extracted to speech. The system reads the text from the Firebase ML provides machine learning capabilities both
image and tells the user about it. It detects text area from online and offline, thus being flexible as required.
natural scene image and extracts text information from the
detected text regions. In text detection, analysis of color 3) Text-To-Speech(TTS): Text-to-speech (TTS) is a process
decomposition and horizontal alignment is performed to of producing spoken word from by converting a given text to
search for image regions of text strings. This method can voice. Text-to-Speech function is a speech synthesizer that
distinguish text in the image from its background. An vocalizes text in real time in a natural way. Android TTS
Adaboost learning model is applied to localize text in camera- library helps to achieve this stage where the extracted text gets
based images. General methods include first, detecting text an audio output in a computer-generated voice. The text
from the image and second, converting the image into speech. received from the text recognizer is analyzed against the
In TEXT DETECTION adjacent grouping methods and text, database for TTS engine and spoken out in an audio format.
character stroke methods configuration is used. In the
extracted text to speech, the mobile speaker informs the user 4) Language Translation: Language translation can be a great
of the text in speech or audio. relief when in a different language speaking place. Several
languages converting APIs are available in the market and can
The mechanical system [5] captures the text from the be opted for the purpose. The extracted text can undergo those
image of a textbook and extract image from the text. The operations to get desired language alterations for better
mechanical set will be consisting of a webcam or maybe an understanding. For language translation, a self- learning
android mobile phone. The captured image is stored in a GUI. statistical machine translation service can be used.
The different image processing techniques used are RGB to a
grayscale image, contrast adjustment, segmented character In this system, based on the analysis of translated texts a
recognition. The segmented character (image) is the input to dictionary is constructed which works as the database of
optical character recognizer which converts it into text. The words and are compared along when a translation is required.
text to speech synthesizer is used to convert text into speech.
Fig. 1 portrays the proposed methodology. The text
III. PROPOSED METHODOLOGY extraction, text to speech and language translation is done
using best available methods.
Our proposed method combines the concept of Optical
Character Recognition (OCR), Text To Speech Synthesizer
(TTS) and language translator. Basically chunking the whole
product into three stages.
The first stage involves extracting text from general
scenes with accuracy using the camera and converting it into
editable text using Optical Character Recognition (OCR). The
second stage involves natural language processing where the
detected text is converted into speech using Text to Speech
synthesizer (TTS). Finally, the detected text is converted to
different languages as per use.
A. Steps of Our Proposed Methodology
Basically the method deals with capturing a textual
image, processing the image to extract text, converting the text
to speech and optionally converting the text to other
languages.
1) Image Capturing: This step involves to extract text from Fig. 1 Proposed Architecture
general scenes with accuracy, mainly from digital images by
means of digital image processing techniques. This is done by
using phone camera. Focused images captured on-the-go or
those stored in the memory space can be utilized in this stage.
IJISRT21MAR439 www.ijisrt.com 566
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. CHALLENGES
Text recognition accuracy is directly proportional to the
clarity and focus of the image. Thus image focus should be of
desirable standards, for suitable results. Also, the image
captured should have its texts of primary coverage and avoid
unnecessary background elements in the image which only
acts as noise for the process.
The way an idea is conveyed via words is different in a
different context and that too different in a different language.
Language translation challenges include this change in con-
textual meaning, as it only translates to the general meaning.
V. CONCLUSION
Textual images and graphical information are in the
scene today. So does their manipulative applications. Text
extraction and manipulation have been in technology for some
years now. Still, no product has gained any noticeable user
satisfaction. By combining the much needed textual image
manipulative operations, our method is an effort to fill that
gap.
REFERENCES
[1]. M. Prabaharan and K. Radha, “Text extraction from
natural scene images and conversion to audio in smart
phone applications,” International Jour- nal of
Innovative Research in Computer and Communication
Engineering, vol. 3, pp. 19–23, 2015.
[2]. J. Yuan, Y. Zhang, K. K. Tan, and T. H. Lee, “Text
extraction from images captured via mobile and digital
devices,” in 2009 IEEE/ASME International
Conference on Advanced Intelligent Mechatronics.
IEEE, 2009, pp. 566–571.
[3]. K. Kalaivani, R. Praveena, V. Anjalipriya, and R.
Srimeena, “Real time implementation of image
recognition and text to speech conversion,” Int. J. Adv.
Res. Technol, vol. 2, pp. 171–175, 2014.
[4]. V. Yadav and N. Ragot, “Text extraction in document
images: highlight on using corner points,” in 2016 12th
IAPR Workshop on Document Analysis Systems
(DAS). IEEE, 2016, pp. 281–286.
[5]. N.-M. Chidiac, P. Damien, and C. Yaacoub, “A robust
algorithm for text extraction from images,” in 2016 39th
International Conference on Telecommunications and
Signal Processing (TSP). IEEE, 2016, pp. 493–497.
