65 nm CMOS Sensors Applied to Mathematically Exact Colorimetric Reconstruction

TL;DR

This paper introduces a mathematically rigorous spectral reconstruction method for colorimetric imaging that is illumination-independent, along with a novel 65nm CMOS sensor design used as a testbed for the approach.

Contribution

It presents a deterministic, least mean squared error spectral reconstruction method and a new CMOS sensor design in standard 65nm technology.

Findings

Reconstruction method achieves accurate spectral approximation.

Sensor design demonstrates high performance in standard CMOS process.

Method is robust to different illumination conditions.

Abstract

Extracting colorimetric image information from the spectral characteristics of image sensors is a key issue in accurate image acquisition. Technically feasible filter/sensor combinations usually do not replicate colorimetric responses with sufficient accuracy to be directly applicable to color representation. A variety of transformations have been proposed in the literature to compensate for this. However, most of those rely on heuristics and/or introduce a reconstruction dependent on the composition of the incoming illumination. In this work, we present a spectral reconstruction method that is independent of illumination and is derived in a mathematically strict way. It provides a deterministic method to arrive at a least mean squared error approximation of a target spectral characteristic from arbitrary sensor response curves. Further, we present a new CMOS sensor design in a standard digital 65nm CMOS technology. Novel circuit techniques are used to achieve performance comparable with much larger-sized specialized photo-CMOS processes. The sensor is utilized as testbed for the spectral reconstruction method.