Flat-field and colour correction for the Raspberry Pi camera module

TL;DR

This paper develops methods to correct vignetting and colour crosstalk in the Raspberry Pi camera module v2, enabling high-quality images with custom optics through real-time and post-processing corrections.

Contribution

It introduces a real-time lens shading table and a post-processing colour unmixing matrix to improve image uniformity and colour accuracy with custom optics.

Findings

Significant reduction in vignetting effects.

Restoration of colour saturation at image edges.

Improved image quality with minimal noise increase.

Abstract

The Raspberry Pi camera module is widely used in open source hardware projects as a low cost camera sensor. However, when the stock lens is removed and replaced with other custom optics the sensor will return a non-uniform background and colour response which hampers the use of this excellent and popular image sensor. This effect is found to be due to the sensor's optical design as well as due to built-in corrections in the GPU firmware, which is optimised for a short focal length lens. In this work we characterise and correct the vignetting and colour crosstalk found in the Raspberry Pi camera module v2, presenting two measures that greatly improve the quality of images using custom optics. First, we use a custom "lens shading table" to correct for vignetting of the image, which can be done in real time in the camera's existing processing pipeline (i.e. the camera's low-latency preview is corrected). The second correction is a colour unmixing matrix, which enables us to reverse the loss in saturation at the edge of the image, though this requires post-processing of the image. With both of these corrections in place, it is possible to obtain uniformly colour-corrected images, at the expense of slightly increased noise at the edges of the image.