numericvision 0.1.0

Detects numeric displays in images using OpenCV

numericvision

A Python 3 package for detecting numeric seven-segment displays in images and running perspective correction on them using OpenCV 4.

detect_box_sequences() first applies a series of filters to the input image to reduce noise, bring out primary contours and bridge gaps between digit segments (see images.apply_filters()). Next, all resulting contours are analyzed by comparing their geometric properties to a set of criteria which define what it means for a contour to resemble a seven-segment digit. Lots of assumptions are being made here (see config.cfg). A contour resembling a seven-segment digit is used to instantiate a Box object. Several boxes lined up in a row make up a box Sequence. Sequences are contained in a Bag.

In other words, raw contours serve as input for an object detection pipeline which produces results in the form of a crawlable object tree.

Installation

$ brew install opencv

$ pip install numericvision

Using the package

from numericvision import detect_box_sequences

box_sequences = detect_box_sequences("image.jpg")

Running from the command line

$ python -m numericvision image.jpg