Perceptual Spatial Uniformity Assessment of Projection Displays with a Calibrated Camera

Spatial uniformity is one of the most important image quality attributes in visual experience of displays. In conventional researches, spatial uniformity was mostly measured with a radiometer and its quality was assessed with non-reference image quality metrics. Cameras are cheaper than radiometers and they can provide accurate relative measurements if they are carefully calibrated. In this paper, we propose and implement a work-flow to use a calibrated camera as a relative acquisition device of intensity to measure the spatial uniformity of projection displays. The camera intensity transfer functions for every projected pixels are recovered, so we can produce multiple levels of linearized non-uniformity on the screen in the purpose of image quality assessment. The experiment results suggest that our work-flow works well. Besides, none of the frequently referred uniformity metrics correlate well with the perceptual results for all types of test images. The spatial non-uniformity is largely masked by the high frequency components in the displayed image content, and we should simulate the human visual system to ignore the nonuniformity that cannot be discriminated by human observers. The simulation can be implemented using models based on contrast sensitivity functions, contrast masking, etc. Introduction In the past decade, tremendous growth in the use of digital media implies that our daily life and work have been greatly impacted by the rapid advancement of display technologies. Hence, the image quality assessment of displays become an essential topic for both scientific research and industrial communities. Projection displays have advantages like high resolution, portability and flexibility. For example, multiple projectors can be tiled up to form a large perceptual photometric seamless image [1]. It is cost effective for users to visualize information in a very high resolution without issuing a customized manufacturing demand. In general, the image quality of displays can be characterized by groups of image quality attributes. One group of them includes physical screen dimension, display resolution, refreshing rate, viewing distance, and viewing angle etc. These attributes are associated with a specific display and its viewing condition. They indeed have impacts on the perceptual image quality, but in most cases they are assumed to be constants within one working cycle of image quality assessment. The rest of the attributes include, but are not limited to, brightness, contrast, color gamut, sharpness and artifacts (including noises). Among these attributes, the spatial uniformity can be of a major importance for projection displays [2, 3]. Researchers tried to achieve objective spatial uniformity with mathematical modeling, but soon they realized that some restraints can be relaxed due to the limitation of perception of Human Visual System (HVS) [1]. In recent studies [4, 5, 6], radiometers were used as absolute acquisition devices to measure the luminance and chrominance of projection displays. Measuring with radiometers is time consuming. The devices are expensive and they are likely to be unavailable in real practice. Digital still cameras have been used to record projection pixels including its background and surrounding on the displays [7, 8, 9]. Cameras have the advantage that they can be placed at different locations in order to achieve a locationand view-dependent image quality assessment, and the acquisition process are much accelerated. However, cameras offer relative sensor responses upon the incoming lights, so they need to be carefully calibrated in advance. In this paper, we propose and implement a work-flow to use a calibrated camera as a relative acquisition device to record the intensity of projections, and evaluate the spatial uniformity of projections against image quality metrics. The correlation between perceived and measured results suggest that the camera based image quality assessment can be reliable and accurate. This paper is organized as follows: first, in the background section, we review the existing image quality metrics for spatial uniformity assessment. Then in uniformity assessment section, we describe the setup of our control lab environment, and demonstrate how to calibrate a camera and a projector to produce multiple levels of linearized non-uniformity on the projection screen. In addition, we also describe the experiment procedure and show the experiment results. In the last section, the conclusions and future works are presented. Background Many uniformity metrics have been proposed based on luminance measurements of gray patches. Among the international standards for image quality assessment of displays, FPDM [10] defines uniformity as (100% · (Lmin/Lmax)), where Lmin and Lmax stand for minimum and maximum measured luminance respectively. TCO 6.0 [11] defines a compliance threshold based on four luminance samples as (Lmax/Lmin), assuming that the minimum luminance is not even close to zero. SPWG [12] defines uniformity as (100% · (Lmax−Lmin)/Lmax) based on thirteen independent luminance measurements. These metrics associate uniformity with Luminance Ratio (LR) between two extreme pixels. However, Tang [13] and Ngai [14] demonstrated that the LR based methods have inaccurate predictions of the non-linear HVS. Tang [13] incorporated the viewing distance d and spatial derivatives s of luminance to define the uniformity as SFA = d ( Lmax+Lmin−2L ) /s, where L stands for the average of measured luminance. Further research from Samuelson et al. [21] quantify the image quality of an illuminated surface with a proposed spatial frequency analysis algorithm incorporating the dif159 22nd Color and Imaging Conference Final Program and Proceedings and 2nd Congress of the International Academy of Digital Pathology ference of Gaussian function, and they suggested that the average magnitudes of luminance contrast within selected spatial frequency bands are related to the lighting quality of the scene represented by the image. Beyond these studies, Ashdown [15] investigated the influence of Luminance Gradient (LG) on the spatial uniformity and they indicated that their results were more correlated to the subjective perceptual ratings than LRs. However, these studies ignore the factor of viewing distance which is important to the uniformity assessment. Meanwhile, other metrics based on statistical analysis and/or color distances in specific color spaces were proposed. Poulin et al. [16] proposed a metric to determine the spatial uniformity as (100%−STDEV (L)), where STDEV (L) stands for the standard deviation (STDEV) of luminance. Thomas et al. [4] used color differences ∆L and ∆C measured with a spectroradiometer. The results suggested that the chromaticity shifts are significant and they should be accounted for. Another statistics based uniformity is defined as the variation