Provided a full stage study of bank serial number recognition system with 99.92% character accuracy and 99.24% Renminbi accuracy.
2018, [22]
Edge detection system from images
Convolutional neural network (CNN), OCR
Preprocessing and post processing is applied with various phases. Testing accuracy is 95% if image taken at a distance of 10m. Model, failed when apply on blurred images.
2018, [23]
Total amount of the bill from daily uses
OpenCV, Tesseract, OCR
This research shown a good result when apply on digital images invoices but often for handwritten bills. Accuracy and efficiency relatively good.
2017, [24]
Printed hard copy documents
OCR
Extracted all kinds of blurred images and converted binary text object into ASCII
2009, [25]
Chinese bank bills, scanned input
Text extraction tools.
Applied preprocessing, image processing and then number and letter segmentation. They got 92% accuracy on letter recognition and 99 % accuracy on number recognition.
2017, [26]
Scanned and camera capture images of ERP system
Stroke Width Transform (SWT) text detection
Extracted the edge of images using the canny edge detector, define the character candidate region, and then clustering the region into text.
2019, [27]
Invoice information based on template matching
OCR along with Contour extraction, edge detection
Identify the information in four steps; preprocessing, matching of template, apply OCR, and exporting of information. They claimed that important information such as money, goods and purchases were accurately identified with 95% of accuracy.
2019, [28]
Visual text of information
CNN based feature extractor
EATEN (Entity-aware attention text extraction network) proposed which is a single shot method to extract information. Study provides a new perspective of text recognition.
2018, [29]
Arabic and Latin scanned invoice
OCR without layout
Entity extraction framework is proposed. Structure relationship graph built for entities. Failed to capture the large portion of Arabic invoices. Work carried out in around 150 real invoices.
2016, [30]
Chinese bank statement images
OCR engine for digits
Geometrical and sentimental knowledge is used to recognize the digit. Showing the page layout accuracy in 98% and recognition accuracy is 89%. Work carried out on 569 bank statement images
Table1: Relevant Studies on Text Extraction from Bills
Tables at a glance
Figures at a glance