Artefact Characterisation for JPEG and JPEG 2000 Image Codecs: Edge Blur and Ringing

Transcription

1 I'.NCINEER- Vol. XXXX, No. 3, pp. 25-3, 27 <D The Institution o Engineers, Sri Lanka Arteact Characterisation or JPEG and JPEG 2 Image Codecs: Edge Blur and Ringing Amal Punchihewa Abstract: This paper presents a method o objective evaluation o the edge blur and ringing arteacts produced by image codecs. It is diicult to detect and measure individual arteacts in coded images. A synthetic test pattern is developed to separate and measure edge blur and ringing arteacts due to image compression. The inluence o contrast on the quality metrics is investigated. The perormance o ive JPEG and JPEG2 codec implementations is compared. All ive codecs show an increasing level o arteacts with increasing compression ratio. Dierent implementations o a codec have dierent arteact characteristics. The objective arteact measures can be used in the image and video codec development process and in parameter optimisation o codec perormance. These metrics can also be used to select suitable parameters or video codecs such as MPEG-, MPEG-2 and MPEG-4 while creating video streams or the Internet applications and in any multimedia application in general. Keywords: Blur, ringing, arteact characterisation, objective assessment, test patterns, arteact metric. Introduction In digital television broadcasting, video streaming and other multimedia communications, image and video are the dominant components. With limited communication bandwidth and storage capacity in terminal devices, it is necessary to reduce data rates using digital codecs. The techniques and quantisation used in image and video compression codecs introduce distortions known as arteacts. Tlie Digital Fact Book deines arteacts as "particular visible eects, which are a direct result o some technical limitation" []. Arteacts are generally not evaluated by traditional methods o signal evaluation. For instance, the visual perception o contouring in a picture cannot be related to signal-to-noise ratio []. High levels o compression result in undesirable spurious eatures and errorpatterns in the reconstructed image; these are the arteacts deined above. Image compression schemes such as JPEG use the techniques o discrete cosine transorm (DCT), block processing and quantisation. JPEG2 uses tiling, wavelet transormation and quantisation. Figure shows some o the resulting arteacts, namely blur, ringing and blockiness in coded images [2]. When the original pattern is not ully known, quantiying these arteacts is diicult. In particular, it is diicult to isolate the individual components o arteacts. Other researchers have developed objective metrics based on nonreerenced or reduced reerence techniques [3,4]. They are good or in-service measurements and as estimates. I the original pattern is unknown it is oten diicult to determine the presence and extent o arteacts. Image codec development, parameter tuning and benchmarking all require availability o more accurate and switer measurements. This can be achieved by using a synthetic pattern having known spatial Edge-blur, ringing and blockiness Figure. Example o Blur and Ringing resulting rom JPEG codec at high compression ratio. Eng. Amal Punchihewa, obtained his B.Sc. Engineering degree with honours, specialising in Electronic and Telecommunication Engineering rom tjie University o Moratuwa, Sri Lanka in 986. He received the Master o Electronics Engineering degree majoring in Digital Video Signal Processing rom the Technical University o Eindhoven, The Netherlands in 99. He was employed as a computer hardware engineer rom 986 to 989 in Sri Lanka. Between 992 to 994, he held a post as the engineer research & planning at the national TV broadcaster in Sri Lanka. Presently, he is a Lecturer at the Institute o Inormation Sciences & Technology at Massey University, Palmerston North campus. He is a Fellow o IESL. 25 ENGINEER

2 distributions o pixel values designed to emphasise the arteacts to be assessed. The study, concentrates primarily on two coding arteacts, namely edge blur and ringing. 2. Objective Quality Metrics Peak signal-to-noise ratio (PSNR) and mean square error (MSB) are commonly used as objective quality measures. They are not good indicators o arteacts present since PSNR and MSB are global pixel error calculations, which do not detect arteacts separately. Yao et al. presented a blurring distortion measure based on the ull reerence approach and computed the correlation between the original and error image edges [5]. However the method does not distinguish blur and ringing which in general may occur concurrently. The author has already developed objective arteact metrics and synthetic test patterns or evaluation o blockiness and colour bleeding arteacts [6, 7]. A search o the literature did not reveal any objective arteact metric or ringing or edge blur using a synthetic test pattern. 3. Methodology The aim o this research was to make use o the concentric circles synthetic test pattern to evaluate the perormance o JEPG and JPEG2 codecs. The test pattern has been designed such that the spatial distribution o pixel values will emphasise edge blur and ringing arteacts due to codec operation [6]. Many image compressors have a control parameter, the quality actor, that can be set by the user to adjust the compression ratio. In general the lower the quality actor the higher the compression ratio and the more visible arteacts become. At low compression ratios, the arteacts may not be obvious or apparent to the human eye. 3. Deinition o arteact metrics Ringing is an undesirable visible eect that always occurs at edges and blur generally occurs at edges. Since I am concerned with the blur occurring at the edge, the rest o this paper concentrates on edge blur. Many codecs transorm the pixel values into the requency domain where the transormed coeicients are then quantised. Quantisation errors resulting rom this approach give rise to ringing around sharp discontinuities in the image. A sharp edge contains components at all requencies. Any change in the relative amplitudes o any o the components will result in ripples in the image with amplitude corresponding to the error. As a result o energy compaction in a codec, many o the high requency components are very small, and get quantised to zero. This loss o high requency components leads to blur in the reconstructed pattern. Ringing Original step edge Figure 2. Ringing and edge blur at an edge o a onedimensional signal. Ringing and edge blur are deined in Figure 2 (Taken rom [6]). I deine the region between the irst crossings on each side o the edge transition as the edge blur region. Outside o this, rom the start o the irst overshoot on each side, the errors are classiied as ringing. To obtain a measure o edge blur, consider the shaded area in Figure 2. The greater the blur, the larger will be the shaded area. By dividing the area by the step height h, a measure o average blur width can be obtained. In a similar manner, the area between the ringing signal and ideal signal provides a measure o the severity o ringing. The measures can be normalised by dividing the average measures by the number o edge pixels N. With sampled data, an ideal step edge would involve a transition between two pixels, as illustrated by the circles in Figure 3. The crosses in Figure 3 are the pixel values near the edge o the reconstructed image rom a codec. The transition involves many pixels and pixel values may oscillate around each level o pixel values. ENGINEER 26

3 The edge blur and ringing arteact metrics may thereore be quantiied as, ISO 6 H 4 I :,. I t i.! ' : Region; Edee B bl'dr * Region o i *. Oirtn'tH, ' Ideal transition edge Step height = h O 6 p O O O Ve---*..*...,, '' i < "~" LiiCi" ] Relative pixel position Figure 3. Edge blur and ringing or one-dimensional sampled data. Circles represent original pixel value and cross represent reconstructed pixel value. I Error,, blur region Edge blur = Nxh Error ringing region Ringing = Nxh In 2-D images, edges may appear at any orientation. Thereore I consider edge blur and ringing perpendicular to the edge under consideration. By summing the separate errors o individual arteacts over the whole test pattern and normalising by the number o edge pixels and step size as given in equations () and (2), I can obtain a measure o edge blur and ringing per edge pixel. () (2) block processing is used, the edges all at dierent places within the blocks. The synthetic test pattern has been designed to emphasise visible edge blur and ringing arteacts. The pixel values and the shape o the pattern have been careully chosen so that the algorithm could detect coding arteacts completely and adequately 4. Results The arteact metrics were evaluated by applying the synthetic test pattern described in the previous section. A JPEG codec was tested at a range o step heights. A set o ive JPEG and JPEG2 codecs were tested at a range o compression ratios. 4. Impact o step height on edge blur and ringing The loss o high requency components results in ringing around discontinuities and a spread o the transition region. The variation o edge blur and ringing were investigated under dierent step heights by evaluating the arteact metrics or a given quality actor o 2 using a JPEG codec. This experiment was repeated or many quality actor values between and. A general observation is that with increasing quality actor, edge blur and ringing demonstrated a similar trend o rapid decrease with increasing step height, and decrease in absolute values over the decreasing quality actors. The Figure 5 and Figure 6 show the metrics evaluated on a JPEG codec at a quality actor o Design o the test signal To test or edge blur and ringing it is necessary to have step edges within the pattern. These should include edges o all orientations in order to detect any orientation sensitivity inherent in the codec. A circular pattern contains edges o i-very orientation. Initially, pixel values o 64 and 92 have been chosen on either side o the boundary, so that ater reconstruction there is adequate amplitude margin to allow or ringing in the reconstructed pattern. To allow or more edges and resulting error pixels, concentric circles have been incorporated (see Figure 4a and 4b). The spacing has been chosen as a prime number (or example, a spacing o 29) so that i Figure 4(a). Original concentric circles synthetic test pattern 27 ENGINEER

4 Figure 4(b). Reconstructed concentric circles synthetic test pattern with blur and ringing arteacts at edges components move rom one quantisation band to another, with consequent dierences in the level o ringing. The low step heights resulted in smaller errors, however, because o quantisation, these errors become more signiicant relative to the step height. Thereore, when normalised, the results become almost random. When averaged over the whole test pattern these results in the large measures seen. At the other end, or very high steps, there is inadequate range available to adequately represent the ringing in reconstructed values. Hence 64 and 92 were set as suitable pixel intensities o the synthetic concentric test pattern J.a O) T3 LLJ 3 2 c t- T V *H*i Step height Figure 5: The variation o Edge blur with step height or a JPEG codec at quality actor o Perormance comparison o JPEG and JPEG2 codecs It is observed that there is a general increasing trend o ringing and edge blur with increasing compression ratio. This is evident in Figures 7 to 2. JPEG codec 2 and codec 3 perorm with a similar objective arteact measure where as codec introduces the same level o arteacts as codec 2 and 3 at much lower compression ratios. Hence or a given degree o arteacts, JPEG codec 2 and 3 could yield higher level o compression (approximately our old increase than JPEG codec ) as shown in Figure 7 and Figure 8. At some compression levels, ringing errors may actually reduce or higher compression depending on exactly where quantisation levels all. ) 'O5 i 5 Step height Figure 6: The variation o ringing with step height or a JPEG codec at quality actor o ~H JPEG codec JPEG codec.7 L~ «JPEG codec :.6 > en.4 T3 LJJ.3. *~~- - ' ' X X v x & 3 ix? 2 Figure 7: Comparison o three JPEG codecs or edge blur against compression ratio. * It was observed that ringing errors are not monotonic because o a threshold eect. As the step height is changed, the dominant ENGINEER 28

5 4 2 o> n U.b n '5> c i 6.4 -t _+_ JPEG codec JPEG codec 2 JPEG codec 3 X ii NVti J. y It ; I V t, % V I 2 "* *-" igure 8: Comparison o three JPEG codecs or ringing against compression ratio. Figure 9 and Figure show the edge blur and ringing resulted rom two JPEG2 codecs. Both implementations are similar in relation to edge blur. At high compression ratio JPEG2 codec perorms better. At maximum compression, the ringing arteact decreases in Figure 8 and Figure because o severe quantisation CD D).4 T: LU JPEG2 codec JPEG2 codec 2 *++-**"'* X"" ' '.',,,, ' t '.* I'igiire 9: Comparison o two JPEG2 codecs or edge blur against compression ratio..2.8 ) '5> C.6 ir.4 JPEG2 codec «JPEG2 codec 2 * *._>.«*+ ' I,,... _* vs Figure : Comparison o two JPEG2 codecs or ringing against compression ratio. i_ Z! CD O) 4 ~O LU.3 PEG codec 2 H JPEG2 codec 2. " * ir***-* '' - X 7 ','',' Figure : Comparison o JPEG and JPEG2 codecs or edge blur against compression ratio..4.2 g>.8 '5> ( ' 6.4 *-=5jr Vv <, - i r t J JPEG codec 2 t JPEG2 codec 2.. * - - h Figure 2: Comparison o JPEG and JPEG2 codecs or edge blur vs. compression ratio. Figure and Figure 2 depict that both JPEG and JPEG2 codecs introduce increasing amount o ringing and edge blur with increasing compression ratio. Two codecs, JPEG and JPEG2 are rom a single codec developer and JPEG2 codec resulted in better picture quality in terms o edge blur and ringing. 5. Conclusions Edge blur and ringing are an undesirable visible eect ound around edges o reconstructed, digitally coded images. In this paper a perormance comparison o JPEG and JPEG2 codecs or edge blur and ringing arteacts was presented using two objective arteact measures or edge blur and ringing. The approach is based on a known test pattern and measurements o the strength o each arteact in the spatial domain. The arteact metrics are a good representation o arteacts and are swit to 29 ENGINEER

6 calculate. The two measures clearly distinguish between the two arteacts. The concentric circles test pattern was designed with knowledge o the speciic mechanisms and weaknesses inherent in transorm coding. The concentric circles test pattern can be used to evaluate edge-blur and ringing produced by either block based or wavelet based codecs. Edge blur and ringing were present in reconstructed pattern rom both types o codecs. For low compression ratios, both JPEG and JPEG2 codecs perorm similarly. Based on the concentric circles test pattern, it is observed that edge blur and ringing decrease with increasing step height. Thus pixel intensities need to be set in the synthetic pattern such that the reconstructed pattern enables measurable quantities or the metric values. The optimised concentric circles test pattern resulted in increases in edge blur and ringing with increasing compression ratio. At higher levels o compression, the loss o ac components o the requency transorms is greater Since video codecs use wavelets or DCT or intra-rame coding, the synthetic test pattern and the two arteact metrics described here can also be applied to evaluate the perormance o video codecs. The concentric circles test pattern is proved to be useul over a wide range o compressions (compression ratio o 2 to 5). The test pattern has been designed and optimized with knowledge o the speciic mechanisms and weaknesses inherent in compression algorithms. At very high compression ratios other arteacts such as blockiness and global blur become prominent. The author intends to perorm urther research to investigate the applicability o the concentric circles test pattern or global blur and to develop a new arteact metric or global blur. Reerences. B. Pank, Tlie Digital Fact Book. Berkshire: Quantel Limited, 22, p A. Punchihewa and D. G. Bailey, Arteacts in Image and Video Systems; Classiication and Mitigation, Proceedings o Image and Vision Computing New Zealand'22, pp , A. Punchihewa, D. G. Bailey and R. M. Hodgson, A Survey o Coded Image and Video Quality Assessment, Proceedings o Image and Vision Computing New Zealand'23, pp , M.C.Q Farias, M. S Moore, J. M Foley, and S. K.Mitra, Detectability and Annoyance o Synthetic Block and Blurry Video Arteacts, Proceedings o XI European Signal Processing Conerence, France, S. Yao, W. Lin, Z. Lu. EP. Ong and M. Etoh, Objective Quality Assessment or Compressed Video, Proceedings o the International Symposium on Circuits and Systems,, PP , A. Punchihewa, D. G. Bailey and R. M. Hodgson, Objective Quality Assessment o Coded Images: The Development o New Quality Metrics, Proceedings o Internet, Telecommunication Conerence, Adelaide, Australia, ppl-6, A. Punchihewa, D. G. Bailey, and R. M. Hodgson, The Development o a Novel Image Quality Metric and a Synthetic Colour Test Image or Objective Quality Assessment o Digitil Codecs, This paper has been accepted or publication by IEEE TENCON 25 Telecommunication Conerence, Melbourne, Australia, 2-23 November 25. ENGINEER 3