Every day on Flatholm island a written record of events on the island is recorded in a logbook. These books go back to the early 1990s and hold some wonderful impressions of life on the island. As a team, we felt that this was something that we would like to capture in our data physicalisation. The challenge was how to get the handwritten records from these wonderful books so that we could analyse the data. Our project partners Yard, jumped to the challenge by identifying an API for the detection of the handwriting.
After researching several different APIs, Yard chose to use Google Cloud’s Vision API for detecting handwriting in images. The API uses a powerful technique known as optical character recognition (OCR). OCR is a technique that converts the images of the logbook text into machine-readable data. OCR attempts to mimic the way our visual system operates, and their detection algorithms are usually powered by neural networks.
Thus far, we’ve seen some pretty decent results. Take a look at the below logbook entry as an example:
We run this entry image through the vision API. The API successfully returned ~95% of words (with correct spellings). Some words however were detected slightly incorrectly.
The word ‘return’ for example, was detected as ‘retum’.
The next step of this process will be to run the detected handwritten words through a spell checker/dictionary, which should improve our accuracy further. And then we can look at doing some sentiment analysis on the text to identify different emotions and feelings from the entries. This could then potentially link up with the work being done with the ambient lighting and maps