Publication of the paper “Automatic Detection and Rectification of Paper Receipts on Smartphones”

Best Path Research has published the results of its latest research on detecting and rectifying paper receipts using a smartphone camera on the arXiv paper repository at:

https://arxiv.org/abs/2303.05763

In this research we developed a novel technique to perform corner detection on potentially deformed objects so as to detect each of the four corners of a paper receipt while ignoring spurious, interfering objects (that might also have corners). Having identified the four unique corners of the receipt, we use a traditional computer vision technique, called a perspective transform, to convert the potentially deformed image into a straight or rectangular image (as if viewed directly overhead, i.e. the so-called “bird’s eye view”).

Having straightened the image in this fashion, the expectation is that all lines in the rectified image are now horizontally aligned, thus making it much easier to perform downstream processing tasks, such as optical character recognition (OCR), or named-entity (NE) detection, on the rectified image. This research and development was conducted primarily by our CEO, Dr. Edward Whittaker, and resulted in a demo iPhone app that integrated the models and techniques described in the paper. The app used a “fine-tuned” object detection model that was then quantized and converted to Tensorflow Lite, to run in real-time on a mid-range iPhone device.

Keywords: Perspective Transform, Tensorflow Lite, iOS, object detection, corner detection, arXiv
Tags:
news_