Contrast enhanced color polarization image fusion

Abstract

Polarization image fusion is commonly employed to improve the quality of images in various fields such as object recognition and detection. There have been limited investigations on color-based polarization image fusion methods, which predominantly rely on pseudo-color approaches and are primarily suited for human perception. We propose a contrast-enhanced method for color polarization image fusion based on local contrast and multiscale decomposition to effectively reflect the multidimensional information and assist subsequent vision tasks. The method created a basic local contrast map from the luminance channel of intensity and grayscale map of the degree of linear polarization (DoLP), then extracted high-frequency features to fuse with the local contrast map utilizing rolling fast guided filtering and least squares. Finally, the local contrast map and high-frequency features of DoLP were combined and injected into the color-intensity image. The quantitative and qualitative evaluation based on nine color image fusion methods and eight blind image assessment measures revealed that the proposed algorithm not only achieved image contrast improvement and effective information fusion, but also preserved the color information.