The clearString Machine Learning Image Recognition system is based on a locally hosted Inception engine.

• To use the system the clearString Extensions pack needs to be installed on the server where clearString is installed.
• As it is a locally installed service, there is no third-party subscription required to use this.

Process overview

The system is designed to classify images by recognising features within them and then what they most likely to be from a set of categories. The Inception model has been taught how to interpret features within an image (for example line and curves, and more complex features built up from them), but it has no knowledge initially of what these are.

To use the system we have to:

1. Create a new 'model' object.
2. Train it with some sample images where we tell it what category each image is in.
3. Use the model - by showing it an image and asking it to classify it - saying which category it thinks it is most likely to be in.

The system can be trained to recognise a wide range of items. How effective it will be depends on the quality of the images used to train it.