Evaluation of tone-mapping algorithms for focal-plane implementation

Abstract

Scenes in the real world may simultaneously contain very bright and very dark regions, caused by different illumination conditions. These scenes contain a wide range of different light intensity values. Attempting to exhibit a picture of such scene on a conventional display device, such as a computer monitor, leads to (a possibly large) loss of details in the displayed scene, since conventional display devices can only represent a limited amount of different light intensity values, which span a smaller range. To mitigate the loss of details, before it is shown on the display device, the picture of the scene must be processed by a tone-mapping algorithm, which maps the original light intensities into the light intensities representable by the display, thereby accommodating the input high dynamic range of values into a smaller range. In this work, a comparison between different tone-mapping algorithms is presented. More specifically, the performances (regarding processing time and overall quality of the processed image) from a digital version of the tone-mapping operator originally proposed by Fern´andez-Berni et al. [11] that is implemented in the focal plane of the camera and from different tone-mapping operators that are originally implemented in software are compared. Furthermore, a second digital version of the focal-plane operator, which simulates a modified version of the original hardware implementation, is considered and its performance is analyzed. The modified hardware implementation is less complex and requires less space than the original implementation and, subjectively, keeps the overall image quality approximately equal to that achieved by digital operators. Issues regarding colors of the tone-mapped images are also addressed, especially the required processing that must be performed by the focal-plane operator after the tone mapping, in order to yield images without color distortions.