This tutorial is aimed at computer vision and deep learning practitioners who would like to have a better understanding of the image formation process, and the image processing steps to convert raw sensor data into a colour image. Additionally, this tutorial will also cover how camera image quality is quantified and how image quality ultimately affects deep learning and computer vision algorithm performance.

At the end of this session, participants will be able to:

-          Understand and describe the fundamentals of camera optics, image sensors, and image signal processing

-          Describe how camera image quality is quantified, both objectively and subjectively

-          Understand how image quality KPIs can be used to quantify the performance limitations of a deep learning/computer vision system

Tutorial Duration: 2 hours

Part 1 (1 hour)

-          Overview of camera technology

-          Optics

-          Image Sensors

-          Image Processing

-          Introduction to Image Quality

-          Subjective vs objective image quality KPIs

-          Image Quality KPIs for deep learning/computer vision

-          Relevant standards (e.g. IEEE P2020, ISO12233, EMVA1288)

Part 2 (1 hour)

-          Relating image quality KPIs to algorithm performance

-          A review of the state of the art/literature

-          Camera mass production - variation in real-world camera performance

-          How to use image quality KPIs to design camera systems

-          Challenges of existing databases and model training approaches

-          Limitations of public datasets

-          Simulated data considerations

-          Dataset generation - pitfalls to avoid

-          On a positive note:

-          How to apply knowledge of camera technology, image quality, and tips for dataset generation to improve system performance