Since last posts, we have gathered approximately 2000 images from 29 objects in the Allard Pierson museum. The reason for collecting that many images is such that the system will be able to recognise the objects from all angles and various distances. Different from our biological ability to capture the visual essence of an object after seeing it only once, computers need more data to perform this task. They simply don’t know what the object will look like when you turn the camera a bit.
By now I can happily announce that we have decided on an algorithm to classify the objects. On our test database this algorithm was able to classify 96% of the images correctly, which is enough for the final App.
Therefore, most of the research is finished, meaning that I can now fully focus on actually building the App!