Raster to vector Ghastly Gallery - Exhibit 1

When we sold raster to vector conversion software we were frequently emailed images to do sample conversions on.

Here is a zoomed in view of part of one such image. It is so ghastly I was inspired to write this article.

This image is too poor quality to convert to CAD using raster to vector conversion software

What is so ghastly about it, you might ask. After all, the lines look quite obvious.

Essentially, there are three issues that make this image so poor - and completely unsuitable for raster to vector conversion:

  1. The text is illegible (e.g. red arrow).

  2. The text and linework touch (e.g. blue arrow).

  3. The background dirt touches the linework and so cannot be easily cleaned away from it - see the image below, which shows a more zoomed view of the area enclosed in the green rectangle:

    Zoomed view of poor quality image

A raster image is made up of pixels that form a shape. The exact nature of this shape is usually obscured by odd pixels that stick out (are these part of the shape or should they be ignored?), by breaks or by other shapes that touch or cross it. As a human being, you can usually unscramble the bits and make sense of it. But all the raster to vector converter sees is a conglomeration of little squares that it somehow has to convert into sensible and correct lines and text characters.

For raster to vector conversion to work you need an image that contains unambiguous shapes. If the image contains a lot of overlapping or touching information, like this one, the shapes become confusing to the raster to vector converter and you will not get a good conversion.

This image is by no means the ghastliest image we have seen, and to be fair parts of the image were better quality than the section shown here.

However, it is not only raster to vector conversion that the image is unsuitable for. Just imagine if it was one of the images in your archive and the original drawings had been destroyed. Although the linework can be made sense of by a human being, the text is illegible and has been irretrievably lost.

I have not seen the original drawing that this was scanned from, but assuming it was legible and reasonably clean there is every possibility that if it had been scanned at a slightly higher resolution using a good adaptive threshold, the raster image quality would have been fine.

I'm hoping that seeing images like this will stress the importance of careful scanning by knowledgable operators, and make companies who pay untrained staff low wages to scan drawings en masse on scanners that are sold on the promise of scanning "400 drawings per hour" think twice.