OPTIMAL TELEVISION SCANNING FORMAT FOR CRT-DISPLAYS Erwin B. Bellers, Ingrid E.J. Heynderickxy, Gerard de Haany, and Inge de Weerdy Philips Research Laboratories, Briarcliff Manor, USA yphilips Research Laboratories, Eindhoven, The Netherlands ABSTRACT At the time of introduction of the television, the video format was optimized given both economical and technological constraints. The resulting video format is not necessarily optimal for the current display technology. Moreover, current consumer-level priced and state-of-the-art scan-rate converters enable a spatio-temporal decoupling of the received video and the displayed video. This paper presents the results of a subjective assessment indicating the preferred CRT-display format. In order to have a fair optimization of the television display format, i.e. comparing options with approximately the same cost 1, we may increase the number of scanning lines at the expense of a lower refresh rate, or a change in the interlace phase. Note that this recently obtained additional freedom enables a variety of pixel distributions in both space and time, for every chosen pixel rate. Section 2 focuses on the conducted experiments and selected environment. Section 3 presents the results of the experiments. Finally, in Section 4 we draw our conclusions. 1 INTRODUCTION Interlaced video with 525 or 625 lines and a 50 or 60 Hz picture rate has been the television broadcast standard for quite some time. Modern bright television screens, however, require a modified display format to prevent annoying large area flicker and/or interline flicker. Moreover, matrix displays require format conversion as they cannot directly cope with interlace. Finally, new image sources, such as the Internet, benefit from an increased spatial pixel density to improve the legibility of displayed textual and graphical information. As such, the optimal scanning or video format requirements may differ per application. The techniques for high quality video format conversion have recently reached a price / performance ratio that enables application in the consumer domain [1, 2, 3]. As a result, we can choose the displayed number of scanning lines, the interlace factor and the picture rate at will. Given this new freedom, the question arises how to optimally choose a display format for the current applications. Increasing the number of scanning lines increases the vertical resolution. Modifying the interlaced scanning to the progressive format eliminates any of the possible interlace artifacts like line flickering and line crawl. Finally, an increase in the refresh rate reduces or eliminates the large area flicker. Consequently, one might expect optimal performance by using both the highest number of scanning lines and at the highest refresh rate possible. Obviously, there is a cost increase associated with such an increase in overall quality. 2 THE EXPERIMENTS A video format is mainly characterized by the spatial resolution, temporal frequency, i.e. update frequency, and the interlace phase. As explained above, a fair optimisation should compare alternatives at equal cost, which implies that we have to look for the right balance from a viewer s point of view between the number of scanning lines, the refresh frequency and the interlace factor at a given pixel frequency. As far as we know, no objective metric yet is able to reliably predict the perceived quality resulting from combinations of these three video format characteristics. Therefore, subjective testing is the only available option for determining the optimal video format. 2.1 Video formats From a viewer s point of view the optimal video format will be the one that offers the best balance between vertical resolution, flickering and annoying interlace artefacts. In those cases that subjects have to balance various aspects of one stimulus (i.e. a still image or a sequence), the pairedcomparison technique is the preferred methodology. It implies that a subject is shown a pair of stimuli, where each stimulus of the pair is displayed in a different video format. The subject is requested to indicate the stimulus that is most preferred. In this way, each subject has to compare for a set 1 the same pixel and line frequency
input format e.g. 625(2:1)@50Hz Video Format Conversion display format 625(1:1)@50Hz 525(1:1)@60Hz 1250(2:1)@50Hz 833(2:1)@75Hz 625(2:1)@100Hz compare As a first format considered in the evaluation we increased the vertical number of lines to arrive at a progressive video signal indicated as 50p (for so-called 50Hzcountries ). Similarly we selected 60p as a viable option. Instead of displaying the 625 lines of the 50p format with the same interlace phase (because it is progressive), we could also consider to display the same amount of lines per field but with a different interlace phase (50i). As such, we create the opportunity to further increase the vertical resolution by a factor of 2! As nice as this might seem, in practice, we have to take the typical interlace artifacts, especially at a refresh rate of 50Hz, into account. And as a result, we need to balance the visibility of interlace artifacts versus the potential double vertical resolution. The visibility of interlace artifacts can be reduced by an increase in the picture refresh rate. However, increasing the refresh rate comes at the cost of lowering the vertical resolution. As such, an compromise is created by the 75i format. Finally, the 100i format is an existing format that is free of visible interlace artifacts and large area flicker. The disadvantage is that the vertical resolution has been reduced to the 625 lines again. Figure 1: Decoupling of the input video format and the display format. of images with various content all mutual combinations of the different video formats in the study. The disadvantage of subjective testing in general is that subjects are only able to assess a limited number of stimuli. Therefore, especially with the paired-comparison methodology the number of video formats that can be evaluated on a selection of image material has to be restricted to about 5 formats. The 5 formats we carefully selected on availability and applicability differ in the number of scanning lines, the interlace phase and the refresh rate, but all use the same pixel frequency (27 MHz 2 ). These formats are (see also Figure 1): 50 Hz, progressive (1:1), 625 scanning lines (50p) 60 Hz, progressive (1:1), 525 scanning lines (60p) 50 Hz, interlaced (2:1), 1250 scanning lines (50i) 75 Hz, interlaced (2:1), 833 scanning lines (75i) 100 Hz, interlaced (2:1), 625 scanning lines (100i) As indicated, all these formats have an intrinsic higher spatio-temporal resolution that the common standard definition video format. As such these formats, therefore, potentially result into an improved appreciation of the standard definition video displayed nowadays, while at the same time the larger sampling frequency has proven to be a viable option. 2 This is twice the pixel frequency for common interlaced video. 2.2 Test set Critical image material, originating from either television cameras, or obtained from Internet pages has been selected for this subjective assessment. In total 4 different still pictures were selected, i.e. Grapes, Text, Siena and Web, as shown in Figure 2. The pictures Siena and Grapes are considered as typical good quality scenes from a television camera, whereas the pictures Text and Web are more typical in a PC like environment or the Internet. Although all scenes contain spatial high frequencies, particularly Text and Web uses the frequency spectrum up to the Nyquist frequency. As we want to compare the various video formats without having to rely on scan-rate conversion quality, we used still picture in these experiments only, i.e. we considered an ideal scan-rate conversion. 2.3 Test conditions The characteristics of the display device often limits the scanning or display format. Moreover, various display types like Cathode Ray Tube (CRT), Plasma Display Panel (PDP), Liquid Crystal Display (LCD), etc, differ in characteristics and do, therefore, not necessarily share the same optimal display format. In our experiments we focused on optimizing the display format for the CRT. Two 21 Philips computer monitor CRTs were used during both sessions. Because of the high vertical resolution that is required to evaluate some of the video formats, computer monitor tubes instead of television tubes had to be
Siena (a) Grapes (b) Text (c) Web Figure 2: Snapshots of the test set. (d)
chosen. This, however, has the disadvantage that the overall luminance of the display is limited, and therefore, the viewing distance had to be fixed to twice (because of the emulated resolution (see Section 2.4)) the standard viewing distance for computer monitor applications, which is 0.5-0.7 m. Thus, subjects assessed all pairs of images at two viewing distances: 1 m and 1.5 m. In the results we did not find a difference between these viewing distances. Therefore, we combined the scores of both. Both CRTs were adjusted to a luminance of 75 cd/m 2 at a maximal contrast and to a colour temperature of about 7000K (7250K for one monitor and 6700 K for the other). In order to prevent that small differences between both monitors affect the results, each pair of images is shown twice, i.e. once with the first image of the pair on the left monitor and the second one on the right monitor, and once vice versa. The environmental illumination was 50 lux. Nineteen (European) subjects ranging in age between 25 and 45 years participated in the experiment. Only part of the subjects are experienced in video processing, whereas the remainder can be considered as non-experts. All have a (corrected-to-normal) visus of at least 1 on the Landolt C-scale. 2.4 Optimizing the spot size A complication, when comparing different scanning formats on a single CRT, is that the spot dimensions cannot be simultaneously optimal for all formats. Clearly a fine spot is required to exploit the highest vertical resolution resulting from the scanning format with the highest number of lines. However, for the scanning formats with a lower line count, such a fine spot may lead to an annoying visibility of the line structure. To prevent the choice of the spot dimensions leading to a bias in the optimization of the scanning format, we optimized the spot size per scanning format as a balance between perceived sharpness and the visibility of line structure in a first session of the subjective test. To realize a variable spot size without changing the display, we emulated a relatively low resolution display on a high resolution monitor. A single line was mapped to a number of scanning lines of the monitor, using a Gaussian filter to represent the spot dimensions of the emulated CRT. As such, we scaled the video formats mentioned before with a factor of two in both dimensions, and increased the viewing distance accordingly. For example, the 50p format was emulated by displaying a 288 lines by 360 pixels picture on a 576 lines by 720 pixel grid. The additional head-room allowed us to emulate a gaussian spot. To arrive at the displayed format, we started by properly downsampling, using a 11 tap FIR filter, the progressively scanned pictures and applied another FIR filter in the linear domain 3 to simulate the ideal gaussian spot. 3 Filtering in the gamma-domain would give rise to a difference in 0.50 0.75 1.00 1.25 1.50 1.75 0.50 97.2 100.0 86.1 80.6 83.3 0.75 2.8 58.3 38.9 38.9 38.9 1.00 0 41.7 36.1 33.3 22.2 1.25 13.9 61.1 63.9 41.7 19.4 1.50 19.4 61.1 66.7 58.3 22.2 1.75 16.7 61.1 77.8 80.6 77.8 (a) 0.75 1.00 1.25 1.50 1.75 2.00 0.75 80.6 63.9 63.9 55.6 44.4 1.00 19.4 63.9 36.1 41.7 41.7 1.25 36.1 36.1 44.4 41.7 38.9 1.50 36.1 63.9 55.6 52.8 47.2 1.75 44.4 58.3 58.3 47.2 55.6 2.00 55.6 58.3 61.1 52.8 44.4 (b) Table 1: Results of the subjective assessment in % for the a) interlaced format, and the b) progressive format with respect to the optimal spot dimension. During the first session of the subjective test subjects assessed for two of the four pictures (i.e. Siena and Text) six different spot sizes in a paired-comparison test. The optimal spot size was determined for the 100 Hz interlaced as well as for the 50 Hz progressive video format, while the optimal dimensions for the remaining formats were deduced from these results. We used a simplified and ideal spot for our emulation, defined by: G(x) = Ce (σ x)2 (1) where C is a constant and σ our width control parameter. For the interlaced video format the σ varied between σ = 0.50, 0.75, 1.00, 1.25, 1.50 and 1.75, whereas for the progressive video format the σ varied in the range σ = 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00. Note that σ = 0 is not equal to a spot of zero width, but reflects the intrinsic spot shape of the CRT. The results of the first session were used to select the best spot size for each of the five video formats described above. During the second session of the subjective test, the various video formats with an optimized spot dimension per format, are evaluated for all four pictures again in a pairedcomparison set-up. 3 RESULTS The results of the first session measuring the subjectively best spot dimension are show in Table 1. The percentages as indicated in the table show the preference of the spot dimension as indicated at the column head over the spot dimension indicated at the row head, e.g. there is a 86.1% preference of spot dimension defined by σ = 1.25 over the one defined by σ = 0.50. A shaded table cell indicates light output for the various filters.
0.50 1.75 1.50 1.25 0.75 1.00 formats 50p 60p 50i 75i 100i 50p 40.8 83.5 97.4 84.2 60p 59.2 84.2 96.7 89.5 50i 16.5 15.8 71.1 44.7 75i 2.6 3.3 28.9 21.0 100i 15.8 10.5 55.3 79.0 Table 2: Overall results of the subjective assessment in %. 0 0.5 1 1.5 2 Quality Scale (a) 60p 50p 50i 100i 75i 0.75 2.00 1.75 1.50 1.25 1.00 0 0.5 1 1.5 2 Quality Scale Figure 4: Overall results of the subjective assessment. 0.4 0.2 0 0.2 0.4 0.6 0.8 Quality Scale (b) Figure 3: Results of the spot optimization for the a) interlaced format, and b) progressive format. that the corresponding preference is significantly different from 50%, which would be the result when subjects were randomly selecting just one of the spot dimensions without having a real preference. These tables of percentages can be transformed towards z-values using Thurstone s law [4]. The resulting z-values are then summed over the columns, resulting in a quality scale value, as shown in Figure 3. The relative distance on the linear quality axis between the various spot sizes as shown in Figure 3 relate to their perceived differences. As the scale is linear, a distance of for example 2 on the quality axis is perceived as twice as strong as a difference of 1 on the same axis. The results for the interlaced format reveals that a σ of 1.00 is perceived as optimal for the preselected list, whereas for the progressive format a slight advantage was found for σ equal to 1.25 over 1.00. As such, a somewhat larger spot seems to be preferred for the interlaced format over the progressive one. The larger spot compensates for the increased line crawl for the 100i format and the non perfect interlacing at 100Hz due to the 50Hz magnetic scatter field. The optimal spot sizes are used in the second session of the experiment eliminating a bias for either of the video formats due to a spot profile that is not equally balanced for the individual formats. The results of this second session view the preferences for the individual video formats, as shown in Table 2. Again, we can use Thurstone s law [4] to compute the z-values, and summed over the columns results in a quality scale value for each video format. The resulting quality scale averaged over all four pictures is given in Figure 4. It shows that the 75i format is the best, followed by the 50i and the 100i formats. The difference in quality between the latter two formats is not statistically significant. The 50p and 60p formats both have the lowest quality, and again, their mutual difference is not statistically significant. The results can be summarized as: 75i > (50i > 100i) > (50p > 60p) where the formats between the same parenthesis belong to the same quality level. There were some small differences in the quality scale values for a given video format amongst the four pictures. The results demonstrated that the preference for the 75i format was most noticed for the pictures that contain highly textured regions (mainly found in graphical or Internet related pictures), which was expected as interlace artifacts start to become most pronounced in highly detailed pictures. Nevertheless, it resulted in the same grouping and ranking of the video formats for each of the four pictures individually. From the subjective evaluation, we draw the conclusion that the video format of 75i is superior to the alternative scannings. A remarkable observation is that the preference
is most impressive in comparing the 75i format with the two progressive display formats (50p and 60p). In fact, all the interlaced formats are preferred over the progressive formats. The loss of vertical resolution of the progressive formats is apparently recognized as the most distinguishing element. Another interesting observation is that the preference of the 75i format over the 100i format indicates that the observers clearly notice the difference in vertical resolution, and probably hardly observe any difference in large area flicker or line flicker. The difference of the 100i with the 50i format was not considered to be significant, i.e. it was found difficult to chose between a picture with significant line / large area flicker and a high vertical resolution, and a picture with a significant lower resolution and no visible interlace artifacts. [3] M. Schu, G. Scheffler, C. Tuschen, and A. Stolze, System on silicon-ic for motion compensated scan rate conversion, picture-in-picture processing, split screen applications and display processing, IEEE Tr. on Consumer Electronics, Vol. 45, August 1999, pp. 842-850. [4] P.G. Engeldrum, Psychometric scaling: a toolkit for imaging systems development, (Imcotek Press, Winchester, 2000). 4 CONCLUSIONS Recent progress in scan-rate conversion technology enables a decoupling of the received video format and the display format. As a result, we can choose the displayed number of scanning lines, the interlace factor and picture rate at will. Due to this new freedom, there is a need to investigate the optimal display format for a given application. In this paper, we evaluated various display formats for a CRT display. As the spot dimension can not be simultaneously optimal for all the display formats, we started with subjectively investigating the best dimension for both an interlaced and a progressive format. The results of this experiment were used in the subjective assessment to determine the optimal display format for a CRT display. Our subjective evaluation revealed that our viewers always preferred interlaced scanning over progressive scanning with the same pixel rate. From the range of picture rates that we tested, the 75i format turned out to provide the best balance between flicker and resolution. Finally, our viewers preferred the 100i high-end television format over the 60p high-end format. Of all tested formats, we conclude that the interlaced format at 75 Hz, 833 lines is optimal for CRT displays with a 27 MHz luminance pixel rate. REFERENCES [1] G. de Haan, J. Kettenis and B.D. Loore, IC for motioncompensated 100 Hz TV with natural-motion moviemode, IEEE Tr. on Consumer Electronics, May 1996, pp. 165-174. [2] G. de Haan, IC for Motion-Compensated Deinterlacing, Noise Reduction, and Picture-Rate Conversion, IEEE Tr. on Consumer Electronics, Vol. 45, August 1999, pp. 617-624. Erwin Bellers was born in Enschede, The Netherlands, on November 14, 1965. He received his B.Sc degree in 1987 from the Technische Hogeschool, his M.Sc degree with distinction from the University of Twente in 1993, and his Ph.D degree from the Delft University of Technology in 2000 with his thesis about de interlacing. In 1993, Erwin joined Philips Research Laboratories in Eindhoven as a Research Scientist. In 2000, Erwin joined Philips Research USA as a Senior Scientist. His major interest is in video enhancement algorithms, including resolution enhancement, motion compensated de interlacing, motion estimation, and motion compensated picture rate up conversion. He received the second price in the 1997 ICCE Outstanding Paper Awards program, and was invited as a co author for the Proceedings of the IEEE. His work has resulted in about 15 patents and patent applications, about 20 papers, and 2 books so far. Erwin is member of the IEEE. Ingrid Heynderickx was born in Sint-Niklaas, Belgium, on the 28th of October 1961. She received from the University of Antwerp her M.Sc degree in Physics in 1983 and her PhD degree in Physics in 1986. In 1987, she joined Philips as a research scientist, and meanwhile worked in different areas of research: optical characteristics of displays, processing of liquid crystalline polymers and the evaluation of the functionality of personal care devices. Since 1999, she is a principal scientist in the Video Processing and Visual Perception group at Philips Research Eindhoven, where she is heading the project on Visual Perception of Display Systems.
Gerard de Haan received the B.Sc., the M.Sc., and the Ph.D. degree from Delft University of Technology in 1977, 1979 and 1992 respectively. He joined Philips Research in 1979. Currently he is a Research Fellow in the group Video Processing & Visual Perception of Philips Research Eindhoven and a Professor at the Eindhoven University of Technology. He has a particular interest in algorithms for motion estimation, scan rate conversion, and image enhancement. His work in these areas has resulted in several books, about 70 papers (www.ics.ele.tue.nl/ dehaan/publications.html), some 50 patents and patent applications, and several commercially available ICs. He was the first place winner in the 1995 ICCE Outstanding Paper Awards program, the second place winner in 1997 and in 1998, and the 1998 recipient of the Gilles Holst Award. The Philips Natural Motion Television concept, based on his PhD-study received the European Innovation Award of the Year 95/96 from the European Imaging and Sound Association. Gerard de Haan is a Senior Member of the IEEE. Inge de Weerd received the B.Sc. degree in Engineering Physics from the Rijswijk Institute of Technology (Technische Hogeschool Rijswijk), the Netherlands, in 1996. In 1999, she received her MSc degree in Technology & Society (focus on Human-Technology Interaction) from the Technical University of Eindhoven. She joined Philips Research Laboratories in Eindhoven in April 2000 as a research scientist in the Video Processing and Visual Perception Group. The emphasis of her work is on perceptual evaluation of image quality.