Authors :
Jaya Krishna Manipatruni; R Gnana Sree; Ranjitha Padakanti; SreePriya Naroju; Bharani Kumar Depuru
Volume/Issue :
Volume 8 - 2023, Issue 9 - September
Google Scholar :
https://bit.ly/3TmGbDi
Scribd :
https://tinyurl.com/4mxbdh36
DOI :
https://doi.org/10.5281/zenodo.8409861
Abstract :
In this study, we delve into the utilization of
PaddleOCR, a readily available tool for optical
character recognition (OCR), in extracting text from
invoices. It is of utmost importance to accurately extract
data from invoices, including information about vendors,
invoice dates, item descriptions, quantities and prices to
effectively manage finances. We achieved this by
leveraging the powerful deep learning models and pre-
trained weights provided by PaddleOCR to process
invoice images and extract the necessary textual details.
Our investigation commences with a
comprehensive analysis of the PaddleOCR framework,
exploring its capabilities and potential for customization.
We explore various techniques aimed at enhancing
image quality and improving OCR accuracy. The
PaddleOCR framework offers advanced functionalities
such as text detection, recognition and layout analysis
that we seamlessly incorporate into our workflow to
accommodate diverse invoice layouts and formats.To train our OCR model effectively, we curate a
meticulously crafted dataset comprising real world
invoice images with varying characteristics. With this
dataset in hand, we fine tune the PaddleOCR model with
a specific focus on enhancing its performance in
extracting text from invoices.
Upon training the model successfully, we evaluate
its performance using an independent test dataset while
measuring key metrics like Character Error Rate (CER)
and Word Error Rate (WER).
Our research strongly confirms the efficacy of the
PaddleOCR powered system in precisely extracting text
from invoices that have different layouts and formats.
Additionally, we conduct a comparison between our
methodology and other OCR techniques, emphasizing
the benefits of PaddleOCR's advanced deep learning
framework.Furthermore, we seamlessly integrate the invoice
text extraction pipeline into a comprehensive automated
system for invoice processing. This integrated system
streamlines the extraction, parsing, and organization of
invoice data, leading to more efficient financial
workflows. We also consider the potential applications of
this technology, including invoice digitization, data
analytics, and process automation, all of which
contribute to significant improvements in operational
efficiency and reduced manual labour in organizations.
In summary, this research demonstrates the
successful use of PaddleOCR for text extraction from
invoices. Our developed pipeline excels in accuracy and
adaptability across various invoice layouts, paving the
way for increased automation in financial management
and document processing.
Keywords :
PaddleOCR, Optical Character Recognition (OCR), Data Extraction, Data Learning Models, Text Detection, Character Error Rate (CER), Word Error Rate (WER), Invoice Digitization, Data Analytics.
In this study, we delve into the utilization of
PaddleOCR, a readily available tool for optical
character recognition (OCR), in extracting text from
invoices. It is of utmost importance to accurately extract
data from invoices, including information about vendors,
invoice dates, item descriptions, quantities and prices to
effectively manage finances. We achieved this by
leveraging the powerful deep learning models and pre-
trained weights provided by PaddleOCR to process
invoice images and extract the necessary textual details.
Our investigation commences with a
comprehensive analysis of the PaddleOCR framework,
exploring its capabilities and potential for customization.
We explore various techniques aimed at enhancing
image quality and improving OCR accuracy. The
PaddleOCR framework offers advanced functionalities
such as text detection, recognition and layout analysis
that we seamlessly incorporate into our workflow to
accommodate diverse invoice layouts and formats.To train our OCR model effectively, we curate a
meticulously crafted dataset comprising real world
invoice images with varying characteristics. With this
dataset in hand, we fine tune the PaddleOCR model with
a specific focus on enhancing its performance in
extracting text from invoices.
Upon training the model successfully, we evaluate
its performance using an independent test dataset while
measuring key metrics like Character Error Rate (CER)
and Word Error Rate (WER).
Our research strongly confirms the efficacy of the
PaddleOCR powered system in precisely extracting text
from invoices that have different layouts and formats.
Additionally, we conduct a comparison between our
methodology and other OCR techniques, emphasizing
the benefits of PaddleOCR's advanced deep learning
framework.Furthermore, we seamlessly integrate the invoice
text extraction pipeline into a comprehensive automated
system for invoice processing. This integrated system
streamlines the extraction, parsing, and organization of
invoice data, leading to more efficient financial
workflows. We also consider the potential applications of
this technology, including invoice digitization, data
analytics, and process automation, all of which
contribute to significant improvements in operational
efficiency and reduced manual labour in organizations.
In summary, this research demonstrates the
successful use of PaddleOCR for text extraction from
invoices. Our developed pipeline excels in accuracy and
adaptability across various invoice layouts, paving the
way for increased automation in financial management
and document processing.
Keywords :
PaddleOCR, Optical Character Recognition (OCR), Data Extraction, Data Learning Models, Text Detection, Character Error Rate (CER), Word Error Rate (WER), Invoice Digitization, Data Analytics.
