Computer Vision Entity Detection

Computer images can store vast amounts of information that is difficult (or impossible) to extract using traditional programming methods. For example, who is in a photograph, how many windows does a building have, how many cars are in a car park or is the weld of acceptable quality?

Checking these things often relies on humans looking at a photograph or drawing and writing down what they can see – associating keywords with the image to allow a future search to be performed that will pull up the precise record in the database.

As you can imagine, human labelling is time-consuming, tedious to perform and very expensive financially. Often including transcript errors and subjective appraisals of what is there. A computer on the other hand, is a great alternative.

Having a large store of images that cannot be easily retrieved can make the images inaccessible for both businesses and individuals. This is where a computer can be used instead.

If the process of object detection could be automated, thousands, if not millions, of images could be processed in a short period of time producing consistent results for a fraction of the cost of employing a human.

Automated image detection begins with training the software on a series of images you want to detect within subsequent images. The training images are labelled to inform the software what the object is (or is not) and are passed through the system a number of times to ensure the detection process is as accurate as it can be made.

For example, ‘this is a good weld’ & ‘this is a bad weld’, ‘this is a car’ & ‘this is a van’, ‘this is Matthew’ & ‘this is Darren’.

After the training phase has been completed the finished software can be used with a new set of images it has yet to see which will be classified as per the training data and automatically generated labels used within a database to allow for easy searching of the images.

Our development team can work with you to analyse the images you want to recognise, and convert these into a training set before finally converting the resulting data into a workable software application that you and your team/customers can use on your own computer.