Contour tracing of binary images (zero ~ background, nonzero ~ object).

Terminology:

A binary image is an image in which the pixel is boolean (represented here using Grey and zero or non-zero pixel values).

All zero-value pixels are part of the "background".

An object is an connected group of non-zero pixels.

A Contour is a trace of an objects' outer or inner edges. Some objects are solid, having no inner contours (consider a filled circle, or letters such as h, s, k and l). Other objects have "holes", also known as inner contours. The letters a and e have one hole while the letter B has two.

After obtaining a Contours structure (via the contours function) the raw traces (Contour type) can be used for further processing or the contours can be filtered by aspects of interest and selectively re-drawn (drawContour) , perhaps used to mask the original image.

About Holes:

In cases where there is only one hole it is uniquely recorded in the Contours structure. Objects with more than one hole record all inner contours in one vector making them hard to extract separately - this is due to the main data structure not being rich enough to record the holes separately. As of writing, this is not seen as an issue because the desired operation, drawContour, can still be achieved. Changing this behavior should be trivial if desired.

Use:

To use this library it is advised that you preprocess the image, including thresholding (ex: otsu on a grey scale image), to obtain a binary image then call:

cs = contours img

The Contours structure can be accessed directly if desired. It includes an Map of all contours (numbered counting from 1) and a vector of the contour sizes (indexed by contour number, zero index is unused/zero).

The algorithm implemented in this module follows the design laid out in 'A Linear-Time Component-Labeling Algorithm Using Contour Tracing Technique' [1].

1

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.95.6330&rep=rep1&type=pdf