Generating your own cascades

We have seen that the key part of object detection within an image is the cascade file. When you open one up, you will notice that it doesn’t seem to be very complicated. The complicated bit is in the encoded data that defines that features within the cascade algorithm that are used to identify the object. This data needs to be discovered through a training step where the relevant OpenCV functions look at images that both have the object in question and images that do not have the object. You need a much larger set of negative images compared to positive images. For example, the author of the blog entry at http:// coding-robin.de/2013/07/22/train-your-own-opencvhaar-classifier.html needed to use a set of 40 positive images and 600 negative images in order to identify…

You're reading a preview of

The Python Book (Digital) - 1 Issue, The Python Book 4th Edition

DiscountMags is a licensed distributor (not a publisher) of the above content and Publication through Zinio LLC. Accordingly, we have no editorial control over the Publications. Any opinions, advice, statements, services, offers or other information or content expressed or made available by third parties, including those made in Publications offered on our website, are those of the respective author(s) or publisher(s) and not of DiscountMags. DiscountMags does not guarantee the accuracy, completeness, truthfulness, or usefulness of all or any portion of any publication or any services or offers made by third parties, nor will we be liable for any loss or damage caused by your reliance on information contained in any Publication, or your use of services offered, or your acceptance of any offers made through the Service or the Publications. For content removal requests, please contact Zinio.