Dynamic Range Enhancing Methods for HDR Imaging and Displays

Abstract

High dynamic range (HDR) of the real-world scenes and human visual perception exceed the capabilities of current imaging and display systems. Most of the existing imaging and display systems cannot acquire or visualize the entire range of light information of the real world, and hence they have limits in providing realistic experience as human eyes can see. That is, they are low dynamic range (LDR) imaging systems. Traditional standards and ecosystem for image and video contents have been developed focusing on the LDR specification. For that reason, HDR imaging systems, image processing techniques and display devices have been in the spotlight to provide more realistic scenes as the human can see the real world. Recently, HDR technologies have gained momentum, and they have been affecting most of the research fields such as digital imaging, image processing, and display. They aim mainly to increase the dynamic range of visual contents when capturing, processing and visualizing the light information of the real world scene.

In this dissertation, dynamic range enhancing methods for imaging and display systems are presented. The first method is an image enhancement algorithm for undesirably illuminated images by using a single exposure, which is a single image high dynamic range (HDR) imaging method. The input image is first decomposed into illumination and reflectance components by employing an edge-preserving smoothing filter. Then the decomposed reflectance component is scaled up to improve the image details in bright areas. The estimated illumination component is locally and globally scaled up and down to generate several illumination images that correspond to certain camera exposure values which are different from the original. These illuminations are respectively combined with the enhanced reflectance to generate virtual multiple LDR images of multi-exposures. Finally, a detail-enhanced and illumination-equalized image is constructed by employing a tone fusion method that blends the multi-exposure images. A method to generate appropriate weighting maps for the proposed pseudo multi-exposure images is also presented. Also, the algorithm is extended to reverse tone mapping method that reconstructs HDR information from an LDR image for visualizing it on HDR displays. The feasibility is proven, and the performance is demonstrated by objective and subjective comparison with the conventional methods. This dynamic range enhancing method for LDR images is expected to be functional for all self-emission type HDR displays.

The second is to increase the dynamic range of liquid crystal display (LCD) which is one of the representative visualization devices. To expand the dynamic range of LCDs, conventional devices have employed localized backlight elements and their driving algorithms concerning local dimming methods. They are designed only to dim the backlight intensities of unit blocks based on the low-level features of the small image areas of input images. However, in this dissertation, human perception-based backlight intensity scaling algorithm is proposed. By keeping or boosting the backlight intensities of salient regions while suppressing in the others, the quality of salient regions and overall contrast are enhanced. Also, power can be saved by backlight intensity reduction in the non-salient areas, without loss of overall quality regarding human visual perception. In this backlight control process, the energy consumption in each backlight element is controlled by a parameter so that the sum of energies of overall backlight elements is kept within a limit. From the experiments, it is shown that the proposed algorithm stretches the dynamic range of backlight intensities (the range of display luminance) extensively. Thus the resulting images are shown to provide wide dynamic range and good image quality without any artifacts. This dynamic range enhancing algorithm can be an effective display system which is limited in energy consumption since it is very energy-efficient. Specifically, it is expected to be a reference of HDR-available luminance dual modulation systems such as high performing HDR LCDs.

Additionally, a wide range of experiments are conducted, and the results are presented to show the scalability and applicability of the HDR algorithm. As a result, it is proved that the proposed single image HDR algorithm can be extended to various HDR imaging methods, and also it can be applied as a pre- and post-processing units for various image processing systems.