Standard image sharpness and noise measurements are insufficient to accurately predict the performance of the machine vision/AI systems used in automotive imaging. This requires information capacity, which combines the effects of sharpness and noise.

We present a new and highly convenient method for measuring information capacity from the familiar slanted edge test pattern that calculates several additional metrics (KPIs), including Noise Power Spectrum (NPS), Noise Equivalent Quanta (NEQ; the number of detected quanta in photon shot noise-limited systems), Ideal Observer SNR (SNRi; the detectability of small objects), and more. All the KPIs can be mapped over the image surface.

Ideally, camera selection should be based primarily on information capacity- important because fewer pixels means faster calculations and reduced power consumption.

We will discuss the meaning of the new measurements and share recent results.