Thomas Wood Computer vision and data science consultant - - PowerPoint PPT Presentation

Thomas Wood Computer vision and data science consultant

www.fastdatascience.com

Past projects in computer vision

• Face recognition
• Pharma pill bottle detection
• Biometric security
• Fingerprint recognition

Face recognition

• Developed face recognition model https://veridiumid.com/vface/
• Scraped training data of >10,000 celebrities from IMDB etc
• Evaluated model in NIST competitions
• Worked on deployment to Android/iOS
• Technology: TensorFlow on GPUs with Convolutional Neural Network (CNN)
• Started from GoogleNet architecture and adjusted parameters and added output

layer

• Model can receive 2 faces and output probability that they are of the same

individual.

Fingerprint recognition

• Worked on fingerprint recognition project at Veridium.
• Originally using Bozorth Fingerprint algorithm (developed by NIST) but moved on to

neural networks.

• https://veridiumid.com/veridium-id

Pharmaceutical pill bottle

• Worked on prototype for US

pharma giant CVS of neural network to detect presence and

• rientation of pill bottle.
• This is for an automatic repeat

prescription application.

Patent on protection of biometric template

• Mather, JF, Othman, A, Streit, S, Dumitran, I & Wood, T (2017). System and method

for biometric protocol standards. US patent 9838388.

• Patent of how biometric templates e.g. from face or fingerprint can be protected

from attackers.

Other deep learning but non-CV projects

• Tesco: prediction of customer orders, currently deployed at

www.tesco.co.uk

• CV-Library: classification of users by CV content. Deployed at

www.cv-library.co.uk/register

• Boehringer Ingelheim (pharma): predicting cost of clinical trials from PDF

documents including tables.

• NHS England, Higher Education England: predicting doctor attrition.