Abstract:Computational color constancy is an important component of Image Signal Processors (ISP) for white balancing in many imaging devices. Recently, deep convolutional neural networks (CNN) have been introduced for color constancy. They achieve prominent performance improvements comparing with those statistics or shallow learning-based methods. However, the need for a large number of training samples, a high computational cost and a huge model size make CNN-based methods unsuitable for deployment on low-resource ISPs for real-time applications. In order to overcome these limitations and to achieve comparable performance to CNN-based methods, an efficient method is defined for selecting the optimal simple statistics-based method (SM) for each image. To this end, we propose a novel ranking-based color constancy method (RCC) that formulates the selection of the optimal SM method as a label ranking problem. RCC designs a specific ranking loss function, and uses a low rank constraint to control the model complexity and a grouped sparse constraint for feature selection. Finally, we apply the RCC model to predict the order of the candidate SM methods for a test image, and then estimate its illumination using the predicted optimal SM method (or fusing the results estimated by the top k SM methods). Comprehensive experiment results show that the proposed RCC outperforms nearly all the shallow learning-based methods and achieves comparable performance to (sometimes even better performance than) deep CNN-based methods with only 1/2000 of the model size and training time. RCC also shows good robustness to limited training samples and good generalization crossing cameras. Furthermore, to remove the dependence on the ground truth illumination, we extend RCC to obtain a novel ranking-based method without ground truth illumination (RCC_NO) that learns the ranking model using simple partial binary preference annotations provided by untrained annotators rather than experts. RCC_NO also achieves better performance than the SM methods and most shallow learning-based methods with low costs of sample collection and illumination measurement.

Deep Neural Models for color discrimination and color constancy

Self-Supervised Learning of Color Constancy

Divergences in color perception between deep neural networks and humans

Color Constancy by Learning to Predict Chromaticity from Luminance

A Retinal Mechanism Inspired Color Constancy Model

Ranking-Based Color Constancy with Limited Training Samples.

Estimating the scene illumination chromaticity by using a neural network

Color Constancy Via Multibranch Deep Probability Network

Computational Colour Constancy Based on Convolutional Neural Networks with a Cross-Level Architecture

Color Equivariant Convolutional Networks

Multi-Domain Learning For Accurate And Few-Shot Color Constancy

Effective Cross-Sensor Color Constancy Using a Dual-Mapping Strategy.

A Color-Opponency Based Biological Model for Color Constancy

Dive into Illuminant Estimation from a Pure Color View

CLCC: Contrastive Learning for Color Constancy

Color Constancy from a Pure Color View.

DeepIlluminance: Contextual Illuminance Estimation via Deep Neural Networks.

Artificial Color Constancy via GoogLeNet with Angular Loss Function

Data Based Color Constancy.

Convolutional Color Constancy

A New Color Constancy Model for Machine Vision