Samu Koskinen: The cameras of the future will produce better images by measuring the spectrum of light

Photo: Hanna Koskinen

In his doctoral dissertation, Samu Koskinen, M.Sc., studied the utilization of spectral data with white balance algorithms. The amount of information is significantly reduced when all naturally occurring hues are modeled using only the relative differences between red, green, and blue colors. Each wavelength has a significance, especially in the area visible to the human eye. This way of thinking enables new tools to improve the performance of current white balance algorithms, especially in challenging conditions.

Today's cameras have a huge number of different algorithms. They aim to make the image look similar to a human experience.

The most relevant algorithm for color reproduction is the white balance algorithm. It is practically the first and biggest color-correcting step in image processing. The color algorithms that follow are often based on decisions made by the white balance algorithm, which multiplies the errors made in the early stages.

In his doctoral dissertation, Master of Science (Technology) Samu Koskinen introduces new tools for use with current machine learning algorithms. The proposed tools are mainly based on spectral data. In this case, color processing is not limited to the traditional RGB color space, but the tools can utilize the full spectrum.

"This approach can be used to create new tools to be used to produce data for AI models, better see gaps in the existing data, and enable the correction of such problems," Koskinen says.

Spectral sensors help in difficult shooting conditions

In his doctoral dissertation, Koskinen presents a completely new way of utilizing small simple spectral sensors. They identify the special characteristics of light sources, which make it possible to find the white point of the light source much more accurately than before.

"This is especially useful in difficult conditions, such as shooting situations where a single color covers almost the entire scene," Koskinen describes.

The tools presented in the dissertation have already been successfully used in the camera product development and partly also in the final products. The development of tools has benefited millions of end users.

Samu Koskinen was born, raised, studied, and worked in Tampere. He is currently working with cameras at Flock Safety Oy, expanding his imaging expertise also outside the scope of his doctoral dissertation.

Public defence on Friday 31 January

Master of Science Samu Koskinen's doctoral dissertation Computational Color Constancy – Using Spectral Domain to Aid Color Constancy for RGB Images will be publicly examined at the Faculty of Information Technology and Communication of the University of Tampere on Friday 31 January 2025 at 12 noon at Tietotalo, hall TB109, Korkeakoulunkatu 1, Tampere. The opponent is Professor Janne Heikkilä from the University of Oulu. Custos is Professor Joni-Kristian Kämäräinen from the Faculty of Information Technology and Communication at the University of Tampere.