Starting as monochrome, bulky cabinets with dim picture tubes, TVs have evolved to large and colorful flat panels with displays that have a high resolution and contrast. As such, they have earned a prominent place in many living rooms and CRT displays have lost their dominant position to LCDs. An important reason for the success of LCDs is their form factor and weight benefit over CRT displays. Especially for large sizes, CRT displays become heavy and bulky. The differences between both display types, however, go beyond appearances. The displays are based on very different principles to create pictures from the incoming TV-signal, and, therefore, the same broadcast can be perceived differently. LCDs did not perform on par with CRT displays with respect to all picture quality attributes at their introduction as consumer TV displays in the mid 1990s. In particular, motion blur on LCD-TVs has been identified as the most prominent shortcoming of LCD-TVs, but also noise is considered more visible on LCDs compared to CRT displays. Consequently, many picture quality improvements for LCDs have been proposed in the literature that reduce their shortcomings. In order to better understand how a display technology affects picture quality, and how picture quality is improved for a certain display technology, properties of picture registration, reconstruction and perception have to be analyzed more closely. In this thesis, a framework is presented that models these three stages of the video chain to enable us to describe static and dynamic perceptual attributes such as motion blur and noise visibility. The framework models the spatial and temporal frequency characteristics of a TV-signal, including its sampling and reconstruction process, and models which frequencies are visible by the Human Visual System (HVS). Also, motion can be analyzed using the framework by describing the motion trajectory that is observed from a picture sequence that contains a time-sequential displacement of (part of) its content. The framework combines this motion description with the motion-tracking capabilities of the HVS and helps to analyze how temporal display properties can result in perceived spatial impairments. The framework has proven useful to demonstrate the perceived differences between pictures on LCD-TVs compared to TVs with a CRT-display. The visibility of motion blur on LCD-TVs has led to several motion blur reduction algorithms for TV applications. Using our framework, we model several of these motion portrayal improvement methods, including Black Frame Insertion (BFI), Gray Frame Insertion (GFI), Scanning Backlight (SBL), Smooth Frame Insertion (SFI), and Motion Compensated Frame Rate Conversion (MC-FRC) for 50 Hz LCDs and evaluate their performance. From this analysis, we conclude that in TV-applications large area flicker has to be considered in the design of motion blur reduction methods and it follows that only MC-FRC and SFI are suitable motion blur reduction techniques for LCD-TVs at broadcast picture rates. Only at picture rates above the large area flicker threshold (e.g at 100 Hz), SBL is best to use for motion blur reduction. In order to obtain the best perceived motion portrayal on LCD-TVs, motion judder and motion blur have to be prevented. From the framework, the conversion of film and video to the highest possible reconstruction rate seems best for obtaining optimal motion portrayal. If computational complexity and physical limitations of a display are taken into account, optimal motion portrayal is obtained using a combination of methods including MC-FRC, SFI and SBL. In addition, it is also necessary to consider motion blur that is originating from the registration process. We, therefore, discriminate between two motion blur sources: camera blur and display blur originating from picture registration and picture reconstruction, respectively. We demonstrate, using our framework, that, for motion compensated 100 Hz LCD-TVs, a motion-portrayal improvement is best obtained by reducing camera blur, because, for these picture rates, the registration process often is the dominant motion blur source. We derive this conclusion by first introducing the TAW as a metric for camera blur, followed by describing a TAW-estimation method that is suitable for TV-applications, and, thirdly, by measuring the TAW on a broad selection of TV-broadcast signals. Camera blur can be reduced with Motion Compensated Inverse Filtering (MCIF). We present two alternatives, namely, non-linear MCIF and trained-filter MCIF. Both methods use an estimate of the TAW and filtering along the local motion vector. The methods differ in their filter design. Non-linear MCIF combines linear sharpness enhancement with clipping while trained-filter MCIF classifies the local structure along the motion vector and applies the MSE-optimal filter that has been obtained through a training process. Both methods have been evaluated and non-linear MCIF has been found most robust to noise, while a trained-filter was found best at recreating picture details. Video noise is considered to be more annoying if a TV-signal is shown on an LCD than when shown on a CRT display. The influence of the display on noise portrayal has been investigated using the framework and was found to not only reside in the display properties, but also in display specific video processing. In particular, when spatial up-scaling is used to convert a low resolution TV-signal for display on high resolution LCDs, the differences in resolution limits, make noise more visible, compared to using CRT-displays. In addition to the identification of the cause for increased noise visibility on LCDs, we present a novel algorithm that shapes the spatial noise characteristics such that the noise spectrum is extended up to the Nyquist limit. Furthermore, we use our framework to reveal that the display properties of LCDs that relate to the temporal aspects of picture reconstruction lead to a motion-dependent blurring. We show that this motion-dependence is reduced together with motion blur reduction and we have investigated if shaping the temporal noise behavior leads to a noise portrayal improvement. Although we found a change in noise behavior, it is not perceived as a clear improvement. In conclusion, we have modeled the registration, addressing, reconstruction and perception of a TV-signal, and used this model to prove that LCDs require display specific video processing such as motion blur reduction and noise diffusion to obtain the best picture quality for TV applications. We also elaborated the required designs.
|Kwalificatie||Doctor in de Filosofie|
|Datum van toekenning||13 dec 2010|
|Plaats van publicatie||Eindhoven|
|Status||Gepubliceerd - 2010|