Open Access
Add to BibSonomyAbstract
We report on the resolution and noise characteristics of handheld and workstation organic light-emitting diode (OLED) displays in comparison with liquid crystal displays (LCDs). The results demonstrate advantages, in terms of sharpness, of handheld OLED displays with modulation transfer function (MTF) values exceeding 0.60 at the Nyquist frequencies. The OLED workstation included in this study exhibits significant signal contamination among adjacent pixels resulting in degraded resolution performance indicated by horizontal and vertical MTF values of 0.13 and 0.24 at the Nyquist frequency. On the other hand, its noise characteristics are superior to the LCD workstation tested. While the noise power spectral (NPS) values of the OLED workstation are 8.0×10−6 mm2 at 1 mm−1, the LCD workstation has NPS values of 2.6×10−5 mm2. Although phone-size OLED displays have superior resolution and noise per pixel, the perceived resolution characteristics at appropriate viewing distances are inferior to tablet-size and workstation LCDs. In addition, our results show some degree of dependency of the resolution and noise on luminance level and viewing orientation. We also found a slightly degraded resolution and increased low-frequency noise at off-normal orientations in the handheld displays.
© 2013 Optical Society of America
1. Introduction
The use of organic light-emitting diode (OLED) displays has increasingly become widespread for applications in television, workstation, and handheld displays. Since OLED displays have emissive pixels unlike liquid crystal displays (LCDs) that need a backlight source, they can provide more effective power saving, wider viewing angle, and thinner design [1–3]. Consequently, OLED displays are replacing LCDs in handheld devices. OLED displays can also provide high image contrast due to lower black-level luminance, more accurate color reproduction, and faster response [4–6]. These benefits are all relevant to imaging devices. Currently, OLED displays are considered for medical applications. OLED workstation has been introduced to image-guided surgery [7], and the use is expected in endoscopy, pathology, and other clinical purposes. Besides, handheld devices positively adopting OLED displays have played important roles in remote image diagnosis for emergency cases or a situation without access to a workstation [8–12]. However, image quality is dependent on the clinical task. Generally, high resolution, low noise, and consistent contrast are desired in radiology. Since recent X-ray image detectors have accomplished small pixel pitches and improved resolution, displays with sufficiently large pixel arrays are needed to exhibit all image information. Noise in X-ray images is mainly due to quantum mottle. Display noise has to be minimal in order to maintain lower radiation doses. While consistent display contrast is needed for accurate image interpretation across devices, fairly high contrast, especially with low minimum luminance, is beneficial in specific cases such as mammography. Other medical specialties including endoscopy, pathology, and dermatology might require accurate color reproducibility with high contrast and fast temporal response.
In this study, we focus on spatial resolution and noise characteristics of OLED displays. While image characteristics in LCDs have been extensively assessed in terms of resolution, noise, luminance, and reflectance [13–18], no report has been published regarding spatial image characteristics in OLED displays although luminance and contrast characteristics have been documented [2, 4–6]. This study demonstrates the characterization of spatial resolution and noise for OLED displays with a range of display sizes and pixel features in comparison with LCDs. This comparison with dedicated medical LCDs would contribute to the understanding of the applicability of OLED displays in radiology. Additionally, the image characteristics are estimated for various luminance levels and viewing angles. Previous studies have reported on the angular response of luminance and contrast in LCDs [19–23] and improved angular performance in OLED displays [2, 4, 6]. However, the angular dependency in terms of resolution and noise and its luminance dependency has not been reported in the literature. Angular dependency is an important factor to ensure displays perform consistently under different viewing conditions. Since the viewing orientation for handheld displays could vary, the resolution and noise variations with angle need to be characterized. In addition, we evaluate sharpness considering the appropriate viewing distance for each display format and compare perceived resolution characteristics.
2. Materials and methods
2.1. Photometric camera
We used a photometric charge-coupled device (CCD) camera (P199F, Westboro Photonics Inc., Ottawa, Canada) to capture the display screen for the resolution and noise measurements. The camera equips a macro lens (NIKON AF Micro-Nikkor 60 mm f/2.8D, Nikon Inc., Tokyo, Japan) and CCD sensors consisted of 1624×1224 elements with 0.0044-mm pixel pitch. The camera is calibrated at the pixel level to luminance in a range from 0.02 to 50,000 cd/m2 with a 12-bit analog-to-digital conversion. Since the measurements were performed at various camera orientations in the following sections, we first measured a modulation transfer function (MTF) of the camera system to examine the angular dependency of the resolution characteristics. We used an optic resolution pattern chart (1951 USAF resolution target, Edmund Optics Inc., Barrington, USA). We set the resolution chart, which was placed on an illuminated film viewer (KLV-5700, HAKUBA Inc., Tokyo, Japan), at a distance of 300 mm away from the camera lens. We captured the resolution chart with the camera at horizontally varying orientations from −10 to 10 degrees with respect to the perpendicular direction as shown in Fig. 1(a). The camera pixel pitch corresponded to about 0.0010 mm on the resolution chart plane. We acquired luminance profiles on the edge of 2×2-mm square cut from the captured images as indicated by a square with red broken line in Fig. 1(b). We localized the edge line at the horizontal center of the camera field of view (FOV) and set the focus on the edge line. The edge profiles were differentiated and fast Fourier transformed to calculate the MTFs at the respective angles.
Fig. 1 Modulation transfer function (MTF) measurement of a photometric camera at various oriented angles. (a) Experiment layout of the camera and a resolution chart. The camera was oriented at varying angles from −10 to 10 degrees with respect to the perpendicular direction to the resolution chart. (b) The captured pattern image with the camera. The edge line of 2×2-mm square cut indicated by a square with red broken line was used for the MTF calculation.
2.2. Display specifications
We used a 3.7-inch phone-size handheld OLED display (OLEDp1: Nexus One, HTC Corp., Taoyuan, Taiwan), a 4.0-inch phone-size handheld OLED display (OLEDp2: Galaxy S, Samsung Corp., Seoul, South Korea), a 25-inch 2-mega-pixels (MP) OLED workstation (OLEDw: PVM-2551MD, Sony Corp., Tokyo, Japan), a 3.5-inch phone-size handheld LCD (LCDp: iPhone4, Apple Inc., CA, USA), a 9.7-inch tablet-size handheld LCD (LCDt: iPad 3rd generation, Apple Inc., CA, USA), and a 21-inch 3-MP LCD workstation (LCDw: R31, Eizo Nanao Corp., Ishikawa, Japan). The display specifications are listed in Table 1. LCDw is mainly used by radiologists and other physicians for primary image diagnosis. LCDt and LCDw adopt in-plane switching (IPS) technology. OLEDp1 and OLEDp2 use active-matrix organic light-emitting diode (AMOLED) displays with PenTile sub-pixel technology, which allocates green (G) sub-pixels interleaved with alternating red (R) and blue (B) sub-pixels. The R-G-B-G layout is iterated and one pixel consists of two sub-pixels of R-G or B-G. The G sub-pixels are mapped by one-to-one correspondence with input signal pixels. The R and B sub-pixels are sub-sampled reconstructing the chroma signal. The luminance is processed using adaptive sub-pixel rendering filters from the input image.
Table 1. Display specifications with minimum and maximum luminance values and luminance ratio. The luminance values were measured at the maximum brightness setting for the handheld displays and LCDw, and the contrast and brightness parameters were set at 50 with a color space: SMTPE-C for OLEDw.
The brightness settings of the handheld displays were fixed at the maximum for each device throughout the experiments. There are no options to change the contrast in the handheld devices. OLEDw has various on-screen display (OSD) settings for contrast, brightness, and color space. Except in noted cases, the parameters were set to contrast: 50, brightness: 50, and color space: SMPTE-C. In addition, the brightness and contrast parameters were changed at the fixed color space as SMPTE-C. We changed the color space parameter from EBU, ITU-709, SMPTE-C, and OFF for fixed contrast and brightness levels. The brightness of LCDw was fixed at maximum and the contrast was set to the digital imaging and communications in medicine (DICOM) grayscale standard display function (GSDF). Table 1 lists the minimum luminance Lmin and maximum luminance Lmax and the luminance ratio LR= Lmax/Lmin at the specific brightness and contrast for each device.
2.3. MTF
We measured MTFs to characterize spatial resolution for the displays listed in Table 1 using a methodology described in [13]. The handheld displays were set at portrait orientation while the workstation displays were set at landscape orientation. We displayed a horizontal or vertical one-pixel-line pattern embedded in a uniform background, respectively for the vertical or horizontal MTF. The horizontal direction corresponds to the red-green-blue (RGB) sub-pixel direction except for LCDt in which the vertical direction corresponds to the RGB direction. The digital driving levels (DDLs) of the line and background were selected from the conditions: line/background= 15/10, 50/30, 60/50, 70/40, and 100/80%. As the exceptions, 50/30% was excluded for OLEDw and the fixed 60/50% was used for LCDt as listed in Table 2. The OSD settings for OLEDw were set at the referential settings as mentioned in 2.2. The displayed screen was captured with the photometric camera at a magnification corresponding to about 9×9 CCD pixels per display pixel as shown in Fig. 2(a). Figure 2(b) shows the captured screen images displaying the horizontal and vertical line-patterns on each display. The distance from the camera to the screen was approximately 300 mm for OLEDp1 and OLEDp2, and the distance was increased in proportion to the pixel pitch for other displays. The camera was first directed perpendicularly to the screen surface, and the screen image was captured at 0-degree viewing angle. At the 0-degree viewing angle and DDL 60/50%, various OSD settings were tested for OLEDw. The brightness and contrast parameters were changed from 50 to 25, 75, and 100 respectively at the color space of SMPTE-C. We also selected respective color space parameters: EBU, ITU-709, SMPTE-C, and OFF at the fixed contrast 50 and brightness 50. Table 2 lists the measured luminance values (cd/m2) at 0-degree angle with the photometric camera, corresponding to the line and background DDLs. The listed luminance values of OLEDw were measured at the referential OSD settings. The luminance values were determined from the averaged pixel values on the one-pixel line and background area in the captured screen images. Next, the camera orientation was horizontally varied in the angle range from −8 to 8 degrees by 2-degree steps with respect to the perpendicular direction for OLEDp1, OLEDp2, LCDp, and LCDt, in a similar way to Fig. 1(a). The displays were rotated by 90 degrees to measure the vertical MTFs.
Table 2. Digital driving levels (DDL) and luminance values of the line (LN) and background (BG) in the displayed pattern at perpendicular viewing angle.
Fig. 2 (a) The experimental setting of the photometric CCD camera capturing the display screen for measuring the display MTFs. The OLEDp1, OLEDp2, LCDp, and LCDt were set by the portrait orientation while OLEDw and LCDw were set by the landscape orientation. (b) The captured screen images displaying the horizontal and vertical line-patterns. The horizontal direction corresponds to the RGB sub-pixel direction except for LCDt in which the vertical direction corresponds to the RGB direction. Orange dot-line squares showing respective one pixel regions and 0.1-mm scale bars are indicated.
The captured line pattern images were subtracted from the background images, which were acquired by capturing uniform patterns with the same setup. After the subtraction, 1200 line profiles in each captured image were averaged vertically (in the line direction). Line spread functions (LSFs) were obtained by normalizing the averaged profiles by the maximum luminance values. Horizontal and vertical MTFs were calculated by fast Fourier transformation of the LSFs. We represent the MTFs as a function of absolute and relative spatial frequency, with relative spatial frequency being equal to the absolute spatial frequency divided by the Nyquist frequency corresponding to the display system. The MTFs as a function of the relative spatial frequency express how much blur is present regardless of pixel pitch. If there is no resolution degradation on the display, the LSF becomes a square wave and the MTF is given by the sinc function,
where u is spatial frequency (mm−1), p is display-pixel pitch (mm), and fN is Nyquist frequency (mm−1) = 1/2p. In addition, we correct the MTFs in terms of appropriate viewing distances and present the measurements in terms of angular spatial frequency expressed in cycles per degree. According to previous reports [24, 25], typical viewing distances VD (mm) are 600, 300, and 200 mm for workstations, tablet-size handhelds, and phone-sized handhelds respectively. The angular spatial frequency (cy/deg) is calculated by the following expression [26]: The image capture procedures were repeated three times with the same camera positioning for each camera orientation and each DDL pattern, and the corresponding three MTFs were averaged. The standard deviations of the three MTFs were calculated for the 60/50% pattern in the perpendicular direction. Student’s t-test was applied to compare the MTFs at 0.4 cy/mm in relative spatial frequency with various DDLs and camera orientations. P-value less than 0.05 was considered statistically significant.2.4. NPS
We measured noise power spectra (NPS) to characterize spatial noise for the displays according to [13, 14, 17]. Uniform patterns with DDLs corresponding to the background levels listed in Table 2 were displayed and each screen was captured with the photometric camera. The OSD settings for OLEDw were set at the referential settings. However, only when the camera orientation was perpendicular to the display screen, various OSD settings were tested for OLEDw displaying 50% DDL pattern as well as the MTF measurements. Next, the camera orientation was changed in the angle range from −8 to 8 degrees relative to the perpendicular viewing direction for OLEDp1, OLEDp2, LCDp, and LCDt. For calculating one-dimensional (1D) horizontal NPS, a region of interest (ROI) of width 512 × height 40 pixels, which positioned at the horizontal center and upper end in the captured image, was selected and the 512-point horizontal profile was acquired with numerical slit (1×40) averaging 40 pixel values to eliminate vertical noise [27, 28]. After being subtracted from the mean value of 512 data in the averaged profile, the profile was processed with a Hanning window, and fast-Fourier transformed. The window processing works to reduce spectral leakage errors [17] in Fourier space, particularly since displays have periodic pixel structures inducing spectral peaks at frequencies in accordance with the integral multiples of the inverse of the sub-pixel pitch. The 1D NPS(u) (mm2) calculation is expressed as follows,
with k= 0, 1, 2, ···, M−1, where M is noise profile data points 512, s is numerical-slit length points 40, x is effective camera-pixel pitch (mm) on captured plane, u= k/(Mx) is spatial frequency (mm−1), L̄ is average luminance (cd/m2) of 512 data, L(kx) is luminance difference (cd/m2) at (kx) from L̄. The 512×40 ROI was moved vertically without overlaps to repeat the NPS calculation as many times as possible and the NPS were averaged. Vertical NPS was calculated in the same way using the horizontal numerical-slit scanning. Furthermore, two-dimensional (2D) NPS(u, v) (mm2) was calculated by a 2D fast Fourier transformation with a Hanning window processing as follows, with k=0, 1, 2, ···, M−1, and l=0, 1, 2, ···, N−1, where x and y are effective camera-pixel pitches (mm) on captured plane, u= k/(Mx) and v= l/(Ny) are spatial frequencies (mm−1), L̄ is average luminance (cd/m2) in ROI with M × N: 256×256 matrices, L(kx, ly) is luminance difference (cd/m2) at (kx, ly) (cd/m2) from L̄. The image capture procedures were carried out three times at the same camera location for each camera orientation and DDL pattern, and the three NPS were averaged. The standard deviations of the three NPS were calculated for the 50% DDL pattern in the perpendicular direction. The NPS variations with DDLs at 2.0 cy/mm and the variations with camera orientations at 0.2 and 2.0 cy/mm were analyzed using Student’s t-test. P-value less than 0.05 was considered statistically significant.3. Results
3.1. Camera resolution
Figure 3 shows the MTFs of the photometric camera at various orientations ranging from −10 to 10 degrees relative to the perpendicular direction. We recognize that the highest MTF at 0 degree is slightly decreased as the angle is tilted up to −10 or 10 degrees. However, since the MTF variation is too small to substantially affect display resolution measurements at least within the angle range from −8 to 8 degrees.
Fig. 3 Modulation transfer functions of the photometric camera at various viewing angles in the range from −10 to 10 degrees.
3.2. Device characterization
All OLED displays have lower minimum luminance and noticeably higher luminance ratios than LCDs as shown in Table 1. The LSFs in Fig. 4 and MTFs as a function of absolute frequency, relative frequency, and angular frequency given by Eq. (2) in Fig. 5 were measured using the pattern with 60% DDL line and 50% DDL background for all displays. Since the LSFs and MTFs of OLEDw were consistent regardless of the OSD settings, Figs. 4 and 5 show the results at contrast 50, brightness 50 and SMPTE-C color space settings. When the vertical line is displayed in OLEDp1 with PenTile technology, the illuminated red or blue sub-pixels are located at only one side of the green sub-pixels as shown in Fig. 2(b). On the other hand, the red or blue sub-pixels on both sides of the green are illuminated in OLEDp2. The illuminated red or blue sub-pixel locations are determined by the respective sub-pixel rendering algorithms to calculate the luminance, and the algorithm of OLEDp1 causes the asymmetric horizontal LSF. However, the LSF difference between OLEDp1 and OLEDp2 does not result in noticeably different MTFs as a function of absolute frequency in Figs. 5(a) and 5(d). The MTFs in Figs. 5(a) and 5(d) show error bars indicating two standard deviations for the three-times measured MTFs. The estimated uncertainty suggests that the MTFs have sufficient accuracy for comparing sharpness among displays. The LSFs of OLEDw present the most widespread profiles among all displays in this study in both horizontal and vertical directions. Although the displays with smaller pixel pitches have the LSFs with smaller widths and higher MTFs in Figs. 5(a) and 5(d), the LSFs of OLEDw have evidently wider widths than the pixel aperture size. The LSFs exhibit non-monotonic distributions that drop once at the border between the center and the next pixels, and then increase at the next pixels before leading undershoots. The MTFs as a function of relative spatial frequency in Figs. 5(b) and 5(e) allow the spatial resolution comparisons regardless of the pixel pitch. In horizontal direction, LCDt and LCDp have MTFs close to the sinc function given by Eq. (1) expressing the ideal MTF, followed by OLEDp1, OLEDp2 and LCDw. In vertical direction, OLEDp2 and LCDt have almost equivalent MTFs to the sinc function, followed by OLEDp1 and LCDp. The MTF of LCDw is slightly decreased from the handheld displays. The MTFs of OLEDw represent the most degraded resolution characteristics in both horizontal and vertical directions. The MTFs as a function of angular frequency in Figs. 5(c) and 5(f) compare resolution characteristics at appropriate estimated viewing distances for the devices. The MTFs of OLEDw and LCDw are relatively increased because of the long viewing distance compared to the MTFs as a function of absolute frequency in Figs. 5(a) and 5(d). Even though the MTFs corrected by shorter viewing distances for the handheld displays are relatively decreased, LCDt has the best resolution characteristics followed by LCDw.
Fig. 5 Modulation transfer functions as a function of absolute frequency, relative frequency, and angular frequency expressed in cycles per degree for the devices tested in this study. (a) and (d) show error bars representing two standard deviations for measured MTFs.
1D and 2D NPS calculated using Eqs. (3) and (4) for all displays in Figs. 6 and 7 exhibit spectral peaks at the integral-multiple frequencies of (1/sub-pixel pitch) based on the prominent periodic pixel and sub-pixel structures shown in Fig. 2(b). The error bars in Fig. 6 representing two standard deviations of the three-times measured NPS suggest sufficient measurement accuracy to compare the noise characteristics among the displays. The OSD settings of OLEDw did not affect the NPS results, except for the contrast setting 25 increasing the NPS slightly. Accordingly, we show the NPS at contrast 50, brightness 50 and SMPTE-C color space settings in Figs. 6 and 7. Figure 6 represents that LCDp and LCDt have the lowest NPS values followed by OLEDp1 and OLEDp2 in horizontal and vertical directions, while OLEDw has higher NPS than the NPS of the handheld displays, and LCDw has the highest NPS in the devices tested. The 2D NPS of OLEDp1, OLEDp2, LCDp, and LCDt in Fig. 7 depict the lower NPS by the darker gray colors. In contrast, the higher NPS of LCDw and OLEDw are emphasized by the brighter gray colors in the 2D NPS.
Fig. 6 One-dimensional noise power spectra for the devices tested in this study. The error bars represent two standard deviations for measured NPS.
3.3. DDL dependency
Figure 8 shows the horizontal LSFs by various DDLs for OLEDp1, OLEDp2, OLEDw and LCDw. The LSFs of OLEDp1 and LCDw indicate no variation with DDLs although the LSFs appear to be noisy at the DDL with LN/BG= 15/10%. In contrast, the LSF of OLEDp2 has undershoots only at the DDL with LN/BG= 70/40%. While the LSFs of OLEDw are independent of DDLs at the center pixel, the off-center peaks and undershoots exhibit subtle relative luminance differences. Subsequently, while the MTFs of OLEDp1 and LCDw have no significant differences (p>0.05) at relative frequency 0.4 cy/mm by DDLs in Fig. 9, OLEDp2 has a higher MTF at DDL 70/40% (p<0.05) compared to the MTFs at other DDLs. Although the MTF increment at the DDL 15/10% in OLEDp2 from the DDL 60/50% is subtle, the difference is statistically significant. In addition, OLEDw has a slightly decreased MTF (p<0.05) at DDL 15/10% and increased MTF (p<0.05) at DDL 100/80% compared to the MTF at 60/50% at 0.4 cy/mm. The vertical LSFs and MTFs also exhibit similar trends to the horizontal results in Figs. 8 and 9.
Fig. 8 Horizontal line spread functions by various digital driving levels for OLEDp1, OLEDp2, OLEDw, and LCDw.
Fig. 9 Horizontal modulation transfer functions as a function of relative spatial frequency by various digital driving levels for OLEDp1, OLEDp2, OLEDw, and LCDw.
Figure 10 shows the LSFs and MTFs for OLEDp2 in both directions by changing the line DDLs against the fixed 40% DDL background. The LSFs and MTFs at the line DDLs between 60 and 68% are similar to the results at the DDLs of LN/BG= 50/30, 60/50, and 100/80% shown in Figs. 8 and 9. At the line DDL 70%, the LSFs have the strongest undershoot peaks resulting in the highest MTFs in the horizontal and vertical directions. As the line DDL is increased from 70 to 100%, the undershoot peaks of the LSFs gradually become less prominent and the MTFs are decreased. However, even at the line DDL 100%, the MTFs are not similar to the results at the DDLs between 60 and 68%, and the MTFs at the line DDLs between 80 and 100 are converged. LCDp has similar tendencies in the MTFs and NPS to the results of LCDw.
Fig. 10 Line spread functions and modulation transfer functions by various line digital driving levels (DDLs) in 40% DDL background for OLEDp2.
Figure 11 shows the horizontal 1D NPS by various DDLs for OLEDp1, OLEDp2, OLEDw and LCDw. The NPS of OLEDp1 and OLEDp2 decreased and converged as the DDLs are increased from 10 to 80%. The NPS at 10, 30, 40, and 80% DDL have statistically significant differences (p<0.05) at 2.0 cy/mm compared to the NPS at 50% DDL except for 80% DDL in OLEDp2. While the NPS of OLEDw and LCDw exhibit no significant variations (p>0.05 at 2.0 cy/mm) with DDLs between 30 and 80%, OLEDw shows increased NPS and LCDw exhibits decreased NPS at 10% DDL (p<0.05 at 2.0 cy/mm in both cases) from the NPS at 50% DDL. The vertical NPS show similar trends than the horizontal NPS for all devices.
Fig. 11 Horizontal 1-dimensional noise power spectra by various digital driving levels for OLEDp1, OLEDp2, OLEDw, and LCDw.
3.4. Angular dependency
Figure 12 shows LSFs at various camera orientations ranging from −8 to 8 degrees for OLEDp1, OLEDp2, and LCDt. The MTFs by the various orientations are shown in Fig. 13. While the horizontal LSFs of OLEDp1 have the widest widths at 8 degrees, the LSFs exhibit the sharpest profiles presenting most distinct sub-pixel separation at −8 degrees. The vertical LSFs have slightly wider tails at −8 and −4 degrees compared to 0-degree profiles. The LSFs of OLEDp2 do not have obvious variations with the orientations although the vertical LSF appers to have slightly wider width at 8 degrees than the LSFs at 0 degree. LCDt also has LSFs with slightly wider widths at 8 degrees particularly in the vertical direction. Consequently, the horizontal MTFs of OLEDp1 are decreased from the highest MTF at 0 degree as the camera orientation is tilted to 8 degrees, with a statistically significant difference (p<0.05) at the relative frequency 0.4 cy/mm. On the other hand, there are no significant differences (p>0.05) among the horizontal MTFs at 0.4 cy/mm in the angle range from 0 to −8 degrees, and among the vertical MTFs from −8 to 8 degrees. Although the vertical MTFs of OLEDp2 are slightly decreased as the camera orientation is tilted at ±8 degrees, the MTF variations are not significant (p>0.05) at 0.4 cy/mm in the respective horizontal and vertical directions. The Student’s t-test suggests a significantly decreased vertical MTF in LCDt at 8 degrees compared to the MTF at 0 degree.
Fig. 12 Line spread functions at various camera orientations from −8 to 8 degrees for OLEDp1, OLEDp2, and LCDt.
Fig. 13 Modulation transfer functions by various camera orientations from −8 to 8 degrees for OLEDp1, OLEDp2, and LCDt.
Figure 14 shows no noise variations for frequencies higher than 1 mm−1 for OLEDp1, OLEDp2, and LCDt in the NPS comparison at 2.0 cy/mm with the various camera orientations except for the comparison between the horizontal NPS at −8 and 0 degrees for LCDt (p<0.05). For frequencies lower than 1 mm−1, the NPS of OLEDp1 is increased at −8 degrees in the horizontal direction (p<0.05 at 0.2 cy/mm) and at −8, 4, and 8 degrees in the vertical direction (p<0.05 at 0.2 cy/mm) compared to the NPS at 0 degree. OLEDp2 has significantly increased horizontal NPS at −8 and 8 degrees and vertical NPS at −8, 2, 4, and 8 degrees compared to the NPS at 0 degree. In addition, LCDt has significantly increased horizontal NPS at −8, −4, 2, 4, and 8 degrees and vertical NPS at 4 and 8 degrees (p<0.05 at 0.2 cy/mm) compared to the NPS at 0 degree. In contrast, the vertical NPS at −4 and −2 degrees are decreased from the NPS at 0 degrees for OLEDp1 and LCDt with significant differences (p<0.05 at 0.2 cy/mm) except that the NPS decrease at −4 degrees for OLEDp1 is not significantly different (p>0.05). The MTFs and NPS for LCDp show similar results to LCDt.
Fig. 14 One-dimensional noise power spectra by various camera orientations from −8 to 8 degrees for OLEDp1, OLEDp2, and LCDt.
4. Discussion
As shown in Table 1, OLED handheld displays and workstation have much higher luminance ratios due to a lower minimum luminance than LCDs. Therefore, OLED displays are well suited for high contrast images with wide dynamic ranges such as mammography, contrast-enhanced computed tomography (CT) and magnetic resonance imaging (MRI), and T2-weighted MRI. Our study suggests that, in general and as expected, displays with smaller pixel pitch have narrower LSFs and higher MTFs as a function of absolute spatial frequency. The horizontal LSFs revealed that the sub-pixel rendering algorithms in PenTile OLED devices are not the same for all devices although the algorithm difference between OLEDp1 and OLEDp2 does not have a substantial influence on the resolution characteristics.
All handheld displays tested in this study have similar pixel pitches and resolution characteristics close to the ideal display. Although LCDw also has comparable resolution characteristics per pixel to the handheld displays in the horizontal direction, the vertical resolution characteristics are inferior. OLEDw has the most degraded resolution characteristics caused by the broad LSFs in both horizontal and vertical directions. The tails in the LSFs indicate signal interference among adjacent pixels. Moreover, the non-monotonic LSFs with off-center peaks and undershoots at neighboring pixels intimate a discontinuous signal diffusion at the pixel borders. Accordingly, we estimate that the signal interference is caused not by optical leakage but by an electrical phenomenon resulting in a parasitic current from driving electronics. The signal interference is distributed over three pixels and leads to the degraded resolution characteristics. Since the OLED workstation has lower noise compared to the LCD workstation, it would be suitable for displaying low contrast images rather than images with high-frequency signals. All handheld displays tested exhibit superior resolution and improved noise characteristics regardless of the display technology compared to the workstation displays. However, since the viewing distances for handheld displays are generally shorter than the distances for workstations, perceived resolution characteristics are different by display formats. The MTFs expressed in angular frequency demonstrate superior resolution of LCDt followed by LCDp and LCDw at the respective viewing distances. While the perceived resolution of OLEDp1 and OLEDp2 is inferior to that of the LCDs in the horizontal direction, the vertical resolution is comparable to the LCD workstation. The resolution and noise results suggest that handheld OLED displays provide more reliable image display for radiologists and other physicians in remote diagnosis and patient consultations. Moreover, the camera MTF data in Fig. 3 suggest that off-normal viewing does not affect the spatial image characteristics of the measuring system at least within the viewing angle range from −8 to 8 degrees.
We did not find resolution variations in OLEDp1 and LCDw with DDLs tested in the experiments, while the resolution characteristics of OLEDp2 varied for specific DDL conditions. Although we found line DDLs between 70 and 100% against the fixed 40% DDL background in OLEDp2 to have the undershoots in the LSFs and increase the MTFs, other DDL conditions might provide variable resolution characteristics as well. The highest MTF at the DDL 70/40% has statistically significant differences from the MTF at the DDL 60/50%. The DDL dependency of the MTFs could be attributed to the sub-pixel rendering algorithm. Since the one-pixel line patterns were displayed with a one-to-one pixel correspondence, the input signals could not be processed or interpolated with the neighboring pixels, using methods such as down-sampling and up-sampling, in displays other than PenTile devices. On the other hand, the input signals are processed using adaptive sub-pixel rendering filters in PenTile devices, so that resolution characteristics can be altered depending on the DDLs. Such resolution differences by DDL might also be seen in other PenTile devices. While the OSD settings in OLEDw do not affect the image resolution characteristics, the DDLs impact slightly the resolution because of luminance changes at the off-center peaks and undershoots in the LSFs.
We also observed noise variations with DDLs as seen in Fig. 11. The sub-pixel rendering in PenTile devices affects the noise characteristics by DDLs. The rendering algorithms in OLEDp1 and OLEDp2 bring noise reduction effects with statistically significant differences at higher DDLs. Such resolution and noise variations with DDLs could cause inconsistent imae interpretation depending on the luminance levels. Further investigation about influences on medical imaging is needed. Although OLEDw has significantly increased NPS at 10% DDL, the NPS might not have sufficient accuracy because of extremely low luminance (0.825 cd/m2) as shown in Table 2 resulting in an overestimated NPS. When LCDw displays 10% DDL, the luminance differences between the pixels (> 4.82 cd/m2) and pixel gaps (≈ 0 cd/m2) are small, and the small noise amplitude possibly results in the reduced NPS.
As the viewing orientation is tilted to −8 or 8 degrees from the normal direction, OLEDp2 and LCDt exhibit one-pixel line profiles with slightly wider widths leading to degraded resolution characteristics. However, the MTF differences are not statistically significant except for the decreased vertical MTF of LCDt at 8 degrees. Since OLEDp1 has asymmetric sub-pixel arrangements in the horizontal direction, the LSF width and shape at −8 degrees differ substantially from the LSF at 8 degrees. Accordingly, the horizontal MTFs are gradually decreased from the highest MTF at 0 degree only when the orientations are tilted to 8 degrees side. The MTF decrease at 8 degrees has a statistical significance. The low-frequency noise can be increased at more tilted orientations, in most handheld device cases tested, with statistically significant differences because the captured images have remarkable luminance differences between the right and left sides in the camera FOV as the camera orientations are tilted. The luminance distributions with stronger trends could result in higher NPS in low frequency range. For frequencies higher than 1 mm−1, the noise characteristics are consistent regardless of the viewing angle for all displays tested. We limited the angle ranges between −8 and 8 degrees because the resolution consistency of the camera is not ensured beyond that range. Our results suggest that it might be beneficial to maintain a perpendicular viewing direction because some displays could provide not only decreased luminance and contrast but also degraded resolution and low-frequency noise at tilted viewing directions.
5. Conclusion
Handheld OLED displays with PenTile sub-pixel design have superior resolution and noise characteristics compared to workstation displays tested in this study. While the OLED workstation has degraded resolution characteristics because of the significant signal contamination among adjacent pixels, the noise characteristics are superior to those of the LCD workstation. Therefore, the OLED workstation would be suitable for displaying low contrast images while the high luminance ratio would also be beneficial for high contrast images with wide dynamic ranges. The resolution comparison at appropriate viewing distances suggests perceived resolution advantages in the tablet-size and workstation LCDs compared to the phone-size OLED displays. We found statistically significant resolution and noise variations with luminance level in handheld OLED displays possibly due to the sub-pixel rendering method. On the other hand, LCD and OLED workstations had consistent image characteristics except at extremely low luminance levels. In addition, we found a slight angular dependency of the resolution and low-frequency noise in the handheld displays. Since such image characteristic variations could lead to inconsistent clinical decisions, the effects of angular viewing habits on diagnostic performance need to be further investigated.
Acknowledgments
AY and CW acknowledge funding by appointments to the Research Participation Program at the CDRH administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. DOE and the U.S. FDA. The mention of commercial products herein is not to be construed as either an actual or implied endorsement of such products by the Department of Health and Human Services. This is a contribution of the FDA and is not subject to copyright.
References and links
1. Y. Kim, J. Kim, Y. Kim, H. Choi, J. Jung, and B. Lee, “Thin-type integral imaging method with an organic light emitting diode panel,” Appl. Opt. 42, 4297–4934 (2008).
2. J. H. Lee, Y. H. Ho, K. Y. Chen, H. Y. Lin, J. H. Fang, S. C. Hsu, J. R. Lin, and M. K. Wei, “Efficiency improvement and image quality of organic light-emitting display by attaching cylindrical microlens arrays,” Opt. Express 16, 2184–2190 (2008).
3. M. A. McCarthy, B. Liu, E. P. Donoghue, I. Kravchenko, D. Y. Kim, F. So, and A. G. Rinzler, “Low-voltage, low-power, organic light-emitting transistors for active matrix displays,” Science 332, 570–573 (2011). [CrossRef] [PubMed]
4. S. H. Cho, Y. W. Song, J. G. Lee, Y. C. Kim, J. H. Lee, J. Ha, J. S. Oh, S. Y. Oh, S. Y. Lee, K. H. Hwang, D. S. Zang, and Y. H. Lee, “Weak-microcavity organic light-emitting diodes with improved light out-coupling,” Opt. Express 16, 12632–12639 (2008). [CrossRef] [PubMed]
5. S. M. Chen, Y. B. Yuan, J. R. Lian, and X. Zhou, “High-efficiency and high-contrast phosphorescent top-emitting organic light-emitting devices with p-type si anodes,” Opt. Express 22, 14644–14649 (2007). [CrossRef]
6. J. H. Lee, X. Zhu, Y. H. Lin, W. K. Choi, T. C. Lin, S. C. Hsu, H. Y. Lin, and S. T. Wu, “High ambient-contrast-ratio display using tandem reflective liquid crystal display and organic light-emitting device,” Opt. Express 13, 9431–9438 (2005). [CrossRef] [PubMed]
7. K. Nishiyama, M. Sano, S. Jinguji, A. Harada, J. Yoshimura, and Y. Fujii, “An organic light-emitting diode display for use in neuroendoscopic surgery in ventricle,” Acta Neurochir. 154, 1523–1525 (2012). [CrossRef]
8. A. E. Flanders, R. H. Wiggins III, and M. E. Gozum, “Handheld computers in radiology,” RadioGraphics 23, 1035–1047 (2003). [CrossRef] [PubMed]
9. R. J. Toomey, J. T. Ryan, M. F. McEntee, M. G. Evanoff, D. P. Chakraborty, D. J. M. J. P. McNulty, E. M. Thomas, and P. C. Brennan, “The diagnostic efficacy of hand-held devices for emergency radiological consultation,” AJR Am. J. Roentgenol. 194, 469–474 (2010). [CrossRef] [PubMed]
10. J. P. McNulty, J. T. Ryan, M. G. Evanoff, and L. A. Rainford, “Flexible image evaluation: iPad versus secondary-class monitors for review of MR spinal emergency cases, a comparative study,” Acad. Radiol. 19, 1023–1028 (2012). [CrossRef] [PubMed]
11. S. John, A. C. C. Poh, T. C. C. Lim, E. H. Y. Chan, and L. R. Chong, “The iPad tablet computer for mobile on-call radiology diagnosis? Auditing discrepancy in CT and MRI reporting,” J. Digital Imaging 25, 628–634 (2012). [CrossRef]
12. A. F. Choudhri, T. M. Carr III, C. P. Ho, J. R. Stone, S. S. B. Gay, and D. L. Lambert, “Handheld device review of abdominal CT for evaluation of acute appendicitis,” J. Digital Imaging 25, 492–496 (2012). [CrossRef]
13. E. Samei, A. Badano, D. Chakraborty, K. Compton, C. Cornelius, K. Corrigan, M. J. Flynn, B. Hemminger, N. Hngiandreou, J. Johnson, D. M. Moxley-Stevens, W. Pavlicek, H. Roehrig, L. Rutz, J. Shepard, R. A. Uzenoff, J. Wang, and C. E. Willis, “Assesment of display performance for medical imaging systems: Executive summary of AAPM TG18 report,” Med. Phys. 32, 1205–1225 (2005). [CrossRef] [PubMed]
14. A. Badano, R. M. Gagne, and R. J. Jennings, “Noise in flat-panel displays with subpixel structure,” Med. Phys. 31, 715–723 (2004). [CrossRef] [PubMed]
15. J. R. S. Saunders and E. Samei, “Resolution and noise measurements of five CRT and LCD medical displays,” Med. Phys. 33, 308–319 (2006). [CrossRef] [PubMed]
16. K. Ichikawa, Y. Kodera, and H. Fujita, “MTF measurement method for medical displays by using a bar-pattern image,” J. Soc. Inf. Disp. 14, 831–837 (2006). [CrossRef]
17. K. Ichikawa, Y. Kodera, A. Nishimura, M. Hasegawa, and N. Kimura, “Analysis method of noise power spectrum for medical monochrome liquid crystal displays,” Radiol. Phys. Technol. 1, 201–207 (2008). [CrossRef] [PubMed]
18. F. Zafar, M. Choi, J. Wang, P. Liu, and A. Badano, “Visual methods for determining ambient illumination conditions when viewing medical images in mobile display devices,” J. Soc. Inf. Disp. 20, 124–132 (2012). [CrossRef]
19. D. H. Fifadara, A. Averbukh, D. S. Channin, and A. Badano, “Effect of viewing angle on luminance and contrast for a five-million-pixel monochrome display and a nine-million-pixel color liquid crystal display,” J. Digital Imaging 17, 264–270 (2004). [CrossRef]
20. S. Hatanaka, J. Morishita, T. Hiwasa, K. Dogomori, F. Toyofuku, M. Ohki, and Y. Higashida, “Comparison of vewing angle and observer performances in different types of liquid-crystal display monitors,” Radiol. Phys. Technol. 2, 166–174 (2009). [CrossRef] [PubMed]
21. A. Badano and D. H. Fifadara, “Goniometric and conoscopic measurements of angular display contrast for one-, three-, five-, and nine-million-pixel medical liquid crystal displays,” Med. Phys. 31, 3452–3460 (2004). [CrossRef]
22. E. Samei and S. L. Wright, “Vewing angle performance of medical liquid crystal displays,” Med. Phys. 33, 645–654 (2006). [CrossRef] [PubMed]
23. G. C. Kagadis, A. Walz-Flannigan, E. A. Krupinski, P. G. Nagy, K. Katsanos, A. Diamantopoulos, and S. G. Langer, “Medical imaging displays and their use in image interpretation,” RadioGraphics 33, 275–291 (2013). [CrossRef] [PubMed]
24. S. P. Prabhu, S. Gandhi, and P. R. Goddard, “Ergonomics of digital imaging,” Br. J. Radiol. 78, 582–586 (2005). [CrossRef] [PubMed]
25. J. T. Norweck, J. A. Seibert, K. P. Andriole, D. A. Clunie, B. H. Curran, M. J. Flynn, E. Krupinski, R. P. Lieto, D. J. Peck, and T. Mian, “ACR-AAPM-SIIM technical standard for electronic practice of medical imaging,” J. Digital Imaging 26, 38–52 (2012). [CrossRef]
26. A. Badano, M. J. Flynn, and J. Kanicki, High-Fidelity Medical Imaging Displays (SPIE Press, 2004).
27. M. B. Williams, P. A. Mangiafico, and P. U. Simoni, “Noise power spectra of images from digital mammography detectors,” Med. Phys. 26, 1279–1293 (1999). [CrossRef] [PubMed]
28. E. Nishimaru, K. Ichikawa, I. Okita, Y. Tomoshige, T. Kurokawa, Y. Nakamura, and M. Suzuki, “Development of a noise reduction filter algorithm for pediatric body images in multidetector CT,” J. Digital Imaging 23, 806–818 (2010). [CrossRef]
