This work presents original algorithms for the measurement of artefacts impairing the quality of digital video sequences. The intended use of these algorithms is the control of the restoration processes performed on the video in advanced monitors for costumer applications. The problem of the artefact measurement at this stage of the processing chain di ffer from the assessment of quality performed in other applications. Quality assessment aimed to the improvement of the encoding operation, for example, can be done using the original sequence
for comparison, and based on the pixel-by-pixel di erences with it. Quality measurements in intermediate stages of the transmission chain of the video, where the sequence is available in compressed form, can employ useful information contained in the bitstream, such as the
frequency distribution of the frame content, the bitrate, the quantisation step and the error rate, all factors related to the global quality. In the proposed application, i. e. at the monitor, the measurements of the frame degradation must instead take place on the decoded numerical
values of the pixels of the sole altered sequence. In addition, these measurement should require a low computational cost, so that they can be used in real time.
In the rst part of this work some of the existing methods for Quality Assessment are briefly overviewed and classi ed based on the chosen approach to the problems. In this overview three main classes of methods are identi ed, namely the methods based on the measurement of speci c frame and video artefacts, the methods measuring the discrepancies between some statistical properties of the pixel distribution or the sequence parameters and ideal models, and the methods processing highly generic measures with trained classi ers. The rst strategy is
deemed the most promising in the intended application, due to the good achieved results with relatively little computation and the possibility to avoid a long and complex training phase. The proposed algorithms are therefore based on the measurement of speci c video artefacts. A second part of the work is devoted to the identi cation of the main potential degradation factors in one of the most recent encoding standard, namely H264. The main aspects of frame degradation, namely blockiness in smooth areas and texture, edge degradation, and blurriness, are identi ed, and their relationship to the encoding options is briefly examined. Based on this brief inspection, two of the most common artefacts of the transmitted video, namely blurriness
and blockiness, are chosen for measurements estimating the picture quality degradation. The devised algorithms integrate measures of the inter-pixel relationships determined by the artefacts with models of human vision to quantify their subjective appearance. For the blurriness measurement two methods are proposed, the fi rst acting selectively on object edges, the second uniformly on the frame surface. In the measurement of the edge blurriness the hierarchical role of each edge is estimated, distinguishing between the marginal edges of the detail and the edges of the main objects of the frame. The former have reduced contrast and short length compared to the edges of the surrounding shapes, and have little e ect on the overall blurriness impression. Conversely, the
state of the latter is the main responsible of the frame quality aspect. The edge blurriness measure is based on the edge width and steepness, corrected with the edge length and the activity of the surrounding scene. This measure of edge blurriness is further corrected with
a measure of the local scene clutter, accounting for the fact that in cluttered scenes the perception of the artefact is reduced. The resulting method yields blurriness measurements in local frame parts. The correlation of this measurements with subjective impression is evaluated
in experimental tests. The two metrics acting uniformly on the frame measure the decrement in perceived contrast and the lack of detail, respectively. Used together, they are e ective in identifying special types
of blurriness resulting in the generation of large areas with few edges and little contrast. These forms of blurriness generally cause a milder degradation of the perceived quality compared to the blurriness caused by encoding. The ability to distinguish among blurriness types and corresponding quality ranges is veri ed in experimental tests. Also the artefacts resulting from block based compression are analysed with a method acting
on the sole edges and another applied to the whole frame. The edge degradation, consisting in an unnatural geometric alteration of the main objects, was measured from the frequency and length of straight edge fractions and the incidence of square corners. A correction procedure is introduced in order to avoid false alarms caused by natural polygonal objects and by the
intrinsic nature of digital pictures. The measure of the blocking artefact on the frame surface, which appears altered by an unnatural grid, is performed with an original solution especially devised for video frames, and aimed to detect the displacement of the synthetic block edges caused by the motion compensation performed in video encoding. On this purpose very sensitive local blockiness
indicators are devised, and corrected with models of the human perception of discontinuities in luminance in order to avoid false alarms. Vision models are further integrated in the computation of a global frame blockiness measure consisting in a weighted sum of local measures on detection points. The metric is tested with respect to its constance on subsequent frames, robusteness to upscaling and correlation with the quality ratings produced in experiments by a group of human observers.
