Blocking Artifact

The block-based hybrid video codec

David R. Balderdash , Fan Zhang , in Intelligent Image and Video Compression (Second Edition), 2021

9.7 In-loop deblocking operations

Blocking artifacts tin can ascend in the hybrid video codec because of move estimation and transform coding of the DFD betoken. Considering of this, deblocking filters accept been used in various forms in standardized video codecs since H.261. They have been proven to improve both objective and subjective performance and have become progressively more sophisticated over time. Wiegand et al. [6] and List et al. [17] provide excellent overviews of the deblocking filter used in H.264/AVC; a summary is given below.

Deblocking filters perform best when they are integrated into the move prediction loop, every bit this eliminates drift and enables benefits for all frame types. The idea of the deblocking operation is illustrated in Fig. ix.18. The edge samples are filtered according to a ready of criteria that relate to the prevailing quantization conditions coupled with estimates of whether the border is real or induced by coding. The absolute differences of pixels about a block edge are commencement computed and so referenced against the probable size of any quantization-induced artifact. If the edge is larger than whatever that might have been induced by coding, then it is nigh probable to be an actual edge and should not be filtered. It has been reported [6] that the incorporation of the deblocking filter saves as much as 10% in bit rate for equivalent quality.

Figure 9.18

Figure 9.18. 1D edge example.

The weather for filtering to be invoked depend on 2 thresholds, α ( QP ) and β ( QP ) , where the 2nd threshold is much smaller than the first. The deblocking process is described in Algorithm 9.5.

Algorithm 9.5

Algorithm 9.5. Deblocking operation.

Case frames from foreman with, and without, a deblocking filter applied are shown in Fig. ix.nineteen. The subjective benefits of the filtering performance can exist clearly seen. What cannot be seen straight are the benefits of the smoothed and ameliorate correlated DFD betoken which volition enable more than efficient transform coding and hence reduced bit rate.

Figure 9.19

Figure 9.19. Illustration of the effect of a deblocking filter. Left: Without deblocking. Right: With deblocking filter applied.

Read full chapter

URL:

https://www.sciencedirect.com/science/commodity/pii/B9780128203538000189

Wavelet Denoising for Image Enhancement

Dong Wei , ... Alan C. Bovik , in Handbook of Image and Video Processing (2nd Edition), 2005

4.2 Blocking Artifacts

Effigy 5 illustrates an example of suppressing blocking artifacts in JPEG-compressed images. Figure v(a) is a role of the original "Lena" image. Figure 5(b) is the same role of a JPEG-compressed version at 0.25 scrap per pixel (bpp), where blocking artifacts are conspicuously visible. The PSNR of the compressed image is 30.4 dB. Figure 5(c) and Fig. 5(d) are the respective parts in the enhanced images via the DWT-based shrinkage and the UDWT-based shrinkage, respectively. The PSNRs of the 2 enhanced images are 31.one dB and 31.iv dB, respectively; i.due east., the UDWT-based shrinkage achieves better objective quality. Although both of them have ameliorate visual quality than the JPEG-compressed i, the artifacts are more completely removed in Fig. v(d) than in Fig. 5(c); i.eastward., the UDWT-based method achieves a better tradeoff between the suppression of coding artifacts and the preservation of image features.

Figure five. Enhancement of a JPEG-compressed "Lena" epitome: (a) a region of the original "Lena" image; (b) JPEG-compressed image; (c) postal service-candy paradigm via the DWT-based method; (d) post-processed prototype via the UDWT-based method.

Read total chapter

URL:

https://world wide web.sciencedirect.com/scientific discipline/article/pii/B9780121197926500735

H.264 video compression

In Digital Video and Hard disk (2d Edition), 2012

Deblocking filter

In MPEG-ii, information technology is a problem that the inverse transform tends to produce discontinuities – blocking artifacts where two 8 ×8 blocks abut. Many MPEG-2 decoders include mail-processing to mitigate the effects of blocking artifacts, but handling after the fact ("out of the loop") is invisible to the encoder.

H.264 standardizes an adaptive, in the loop deblocking filter. The filter adapts to picture content, such as edges (which typically cause the worst artifacts). Deblocking is standardized, and it takes place within the encoder'due south prediction loop.

Read total chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780123919267500485

Smoothlet Transform

Agnieszka Lisowska , in Advances in Imaging and Electron Physics, 2013

three.2 Postprocessing

In all algorithms based on a quadtree partition or a similar sectionalisation, in that location is the problem of blocking artifacts that are present equally a issue of such a segmentation. Many algorithms have been proposed to reduce the quality degradation along the blocks. They are used as a postprocessing step of the image obtained every bit a result of an approximation. From that, it follows that they are compatible with many coding algorithms, including the one based on the smoothlet transform. The best-known postprocessing methods are the ones based on the wavelet filtering ( Mallat & Zhong 1992), Markov Random Fields (MRF) models (Meier, Ngan, & Grebbin 1999), or Discrete Cosine Transform (DCT) domain (Popovici & Withers 2007).

In this chapter, the simplest possible postprocessing method is used because the assumption is made that smoothlets bargain with rather smooth images. The 2nd reason is that postprocessing is a marginal issue. It is applied only to evidence that the results of smoothlet approximation can be further improved.

The postprocessing method used here is based on the border pixel averaging of ii next blocks. Presume that c 1 and c 2 denote the color values of adjacent pixels from the left (up) and right (bottom) subdomains. Assume that c i c 2 . Then the new pixel colors are computed every bit

(24) c 1 due north = c 1 + c 2 c one 3 , c ii n = c 2 c 2 c i 3 .

In other words, the method works in such a fashion that two adjacent pixels (the beginning one from the commencement subdomain and the second one from the second subdomain) are averaged. It is also possible to apply 1-pixel averaging or more than two-pixel averaging. Nonetheless, as follows from the performed experiments, the method given by Eq. (24) mostly assures the best visual results. In Effigy ii.11, the sample "Bird" segment is presented with and without postprocessing. As one can come across, the postprocessing improved the overall visual quality. Note, however, that images with many details can exist blurred by this method.

Figure 2.xi. The effect of image postprocessing. (a) The zoomed segment of image "Bird" without postprocessing, P South Northward R = 31.21 dB ; (b) the zoomed segment of image "Bird" with postprocessing, P Due south N R = 31.65 dB .

To examination how much postprocessing improves the quality of approximated images, the plots of dependency between the number of smoothlets and the MSE were generated for the smoothlet approximation with and without postprocessing for the tested images. Some of the plots are presented in Effigy 2.12. As one can meet from these plots, the comeback is rather modest and depends on an image. Commonly, it is about a few pct points. The shapes of the plots for the other tested images are very like to the one from Effigy 2.12(a). Of course, more sophisticated methods will lead to more visible quality improvement.

Figure 2.12. The plots of dependency between the number of smoothlets used in the approximation and MSE, with and without postprocessing, for two images: (a) "Bird"; (b) "Monarch." (For colour version of this figure, the reader is referred to the online version of this volume.)

Read full chapter

URL:

https://world wide web.sciencedirect.com/science/article/pii/B9780124077010000029

MPEG-4, H.264/AVC, and MPEG-7: New Standards for the Digital Video Industry

Berna Erol , ... Gary Sullivan , in Handbook of Image and Video Processing (Second Edition), 2005

3.2.1-4 In-Loop Deblocking Filter.

Cake-based prediction and transform coding, including quantization of transform coefficients, can lead to visible and subjectively objectionable changes in intensity at coded block boundaries, referred to as blocking artifacts. In previous video codecs, such as MPEG-ii, the visibility of these artifacts could optionally exist reduced to ameliorate subjective quality past applying a deblocking filter afterward decoding and prior to display. The goal of such a filter is to reduce the visibility of subjectively annoying artifacts while fugitive excessive smoothing that would result in loss of item in the paradigm. However, because the process was optional for decoders and was not normatively specified in these standards, the filtering was required to take place outside of the motion-bounty loop, and large blocking artifacts would still exist in reference pictures that would exist used for predicting subsequent pictures. This reduced the effectiveness of the move-compensation process and allowed blocking artifacts to exist propagated into the interior of subsequent movement-compensated blocks, causing the subsequent movie predictions to be less effective and making the removal of the artifacts past filtering more challenging.

To improve upon this, H.264/AVC defines a normative deblocking filtering process. This process is performed identically in both the encoder and decoder to maintain an identical prepare of reference pictures. This filtering leads to both objective and subjective improvements in quality, shown in Fig. 17, due to the improved prediction and the reduction in visible blocking artifacts. Note that the second version of the ITU-T H.263 standard also included an in-loop deblocking filter equally an optional feature, just this feature was not supported in the most widely deployed baseline profile of that standard, and the design had issues with inverse-transform rounding error effects.

FIGURE 17. A decoded frame of the sequence Foreman (a) without the in-loop deblocking filtering practical and (b) with the in-loop deblocking filtering (the original sequence Foreman is courtesy of Siemens AG) (meet color insert).

The deblocking filter defined in H.264/AVC operates on the 4 × four block transform grid. Both luma and chroma samples are filtered. The filter is highly adaptive to remove as many artifacts as possible without excessive smoothing. For the line of samples across each horizontal or vertical block edge (such as that illustrated in Fig. eighteen), a filtering forcefulness parameter is determined based on the coding parameters on both sides of the edge. When the coding parameters bespeak that large artifacts are more than likely to be generated (e.g., intra prediction or coding of non-naught transform coefficients), larger strength values are assigned. This results in stronger filtering being applied. Additionally, sample values (i.e., a 0-a 3 and b 0-b three, in Fig. 18) along each line of samples to exist (potentially) filtered are checked against several conditions that are based on the quantization stride size used on either side of the edge in order to distinguish between discontinuities that are introduced past quantization, and those that are truthful edges that should not be filtered to avert loss of detail [26]. For example, in the example shown in Fig. 18, a meaning lack of smoothness can be detected between the sample values on each side of the block edge. When the quantization pace size is large, this lack of smoothness is considered an undesirable artifact and it is filtered. When the quantization step size is minor, the lack of smoothness is considered to be the issue of actual details in the scene beingness depicted past the video and it is not contradistinct.

Effigy 18. Example of an border contour. Notations a 0, a 1, a ii, and b 0, b 1, b 2, b 3 correspond sample values on each side of the edge.

Read full chapter

URL:

https://www.sciencedirect.com/scientific discipline/commodity/pii/B9780121197926501133

Basic Motion Estimation Techniques

Mohammed Ebrahim Al-Mualla , ... David R. Bull , in Video Coding for Mobile Communications, 2002

4.half-dozen.half dozen Overlapped Motility Compensation

As already discussed, the BMA assumes that each block of pels moves with a uniform translational motion. Because this supposition does not always hold true, the method is known to produce blocking artefacts in the reconstructed frames. One method that reduces this effect is overlapped motion compensation (OMC). The method was beginning proposed past Watanabe and Singhal in 1991 [101]. In BMA, the estimated block motility vector is used to copy a displaced N × N block from the reference frame to the current N × Due north block in the electric current frame. In OMC, nonetheless, the estimated cake motion vector is used to copy a larger block (say, 2N × iiN) from the reference frame to a position centered effectually the electric current N × North block. As illustrated in Figure 4.8, since they are larger than the compensated blocks, the copied blocks overlap, hence the name overlapped motion bounty. Each copied block is weighted by a smooth window, with higher weights at the eye and lower weights toward the borders. This means that the estimated move vector is given more than influence in the centre of the block, and this influence decays toward the borders, where neighboring motion vectors commencement taking over. This ensures a smoothen transition between blocks and therefore reduces blocking artefacts. Overlapped motion estimation and compensation can also be implemented in the frequency domain, as proposed by Young and Kingsbury [95].

Figure four.viii. Overlapped motility compensation for the top-left quadrant of the electric current block

Another view of the OMC process is that each pel in the current Northward × North block is compensated using more than one motion vector. For example, in Figure 4.8, each pel is compensated using 4 motility vectors. The ready of motion vectors is decided according to the spatial position of the pel within the block. A pel in the acme-left quadrant of the current block will be compensated using the motion vector of the block itself, plus the motion vectors of the blocks to the left of, above, and above left of the electric current cake. Each vector provides a prediction for the pel, and those four predictions are weighted co-ordinate to the spatial position of the pel within the cake. For example, equally the spatial position of the pel gets closer to the left border of the block, a college weight is given to the prediction provided by the movement vector of the cake to the left.

Orchard et al. [102, 103] used this view to codify OMC every bit a linear estimator of the form

(4.32) f ˆ t ( south ) = d n north ( s ) w n ( s ) f t Δ t ( s d n ) ,

where n (southward) = {d due north (s)} is the set of motion vectors used to recoup the pel at location s and wdue north (s) is the weight given to the prediction provided by vector d n . Using this formulation, they solve ii optimization issues: overlapped-motion compensation and overlapped-motility interpretation. Given the prepare of motion vectors n ( s ) estimated past the encoder, they advise a method for designing optimal windows, due westn (south), to exist used at the decoder for motion compensation. Besides, given a fixed window that volition be used at the decoder, they propose a method for finding the optimal set up of motility vectors at the encoder. Note that the latter problem is much more complex than the BMA, since in this instance the estimated movement vectors are interdependent. For this reason, their proposed method is based on an iterative procedure. A number of methods have been proposed to convalesce this complexity, e.g., Ref. 104.

As a linear estimator of intensities, OMC belongs to a more general gear up of move compensation methods called multihypothesis motion compensation. Some other fellow member in this set is bidirectional motion bounty. The theoretical motivations for such methods were presented by Sullivan in 1993 [105]. Recently, Girod [106] analyzed the rate-baloney efficiency of such methods and provided performance bounds and comparisons with single-hypothesis motion compensation (due east.yard., the BMA).

Figure 4.9 compares the functioning of OMC to that of the BMA when applied to the Foreman sequence. In the case of OMC, the same BMA motion vectors were used for compensation (i.e., the motion vectors were not optimized for overlapped compensation). Each motion vector was used to copy a 32 × 32 block from the reference frame and center it effectually the current 16 × sixteen block in the electric current frame. Each copied block was weighted by a bilinear window function defined as [103]

Effigy iv.9. Comparison between OMC and BMA

(iv.33) w ( x , y ) = due west x w y , where w z = { one 16 ( z + i ii ) for z = 0 , , fifteen , w 31 z for z = xvi , , 31.

Edge blocks were handled by assuming "phantom" blocks outside the frame boundary with movement vectors equal to those of the border blocks. Despite the fact that the estimated vectors, the window shape, and the overlapping weights were not optimized for overlapped bounty, OMC provided better objective (Figure four.9(a)) and subjective (Figures 4.9(b)– four.9(d)) quality compared to the BMA. In detail, the annoying blocking artefacts have conspicuously been reduced.

Read total affiliate

URL:

https://www.sciencedirect.com/scientific discipline/article/pii/B9780120530793500068

Filter-banks and wavelet compression

David R. Bull , Fan Zhang , in Intelligent Image and Video Compression (Second Edition), 2021

six.6.ane Overview

JPEG has been a big success, with around lxxx% of all images still stored in this format. However, in the late 1990s, JPEG'south express coding efficiency, the presence of annoying visual blocking artifacts at high compression ratios, limited colour gamut, and limited resolution were all reasons why work on an improved standard began. Furthermore, many applications were emerging that required functionalities not supported past JPEG, for instance spatial scalability, SNR scalability, and region of involvement (ROI) coding.

The aim of JPEG2000 was to reach a thirty% bit rate saving for the same quality compared to JPEG, supporting 2 to xvi 1000000 colors and both lossless and lossy pinch within the same architecture at bit depths greater than 8 bits. Additional features to support ROI coding, fault resilience, and information security were likewise included and a high accent was placed on scalability and progressive manual. JPEG2000 Part 1 became an International Standard (ISO/IEC 15444-1) in December 2000 [21,29,31].

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780128203538000153

Iterative Image Restoration

Aggelos K. Katsaggelos , ... Chun-Jen Tsai , in The Essential Guide to Epitome Processing, 2009

15.half dozen.iv.1 Removal of Compression Artifacts

The problem of removing pinch artifacts addresses the recovery of information lost due to the quantization of parameters during pinch. More specifically, in the bulk of existing image and video pinch algorithms the image (or frame in an image sequence) is divided into foursquare blocks which are candy independently from each other. The Discrete Cosine Transform (DCT) of such blocks (representing either the image intensity when dealing with nevertheless images or intracoding of video blocks or frames, or the displaced frame deviation when dealing with intercoding of video blocks or frames) is taken and the resulting DCT coefficients are quantized. As a effect of this processing, annoying blocking artifacts outcome, primarily at high compression ratios. A number of techniques have been developed for removing such blocking artifacts for both still images and video. For case, in [ 27, 28] the problem of removing the blocking artifacts is formulated as a recovery trouble, according to which an gauge of the blocking antiquity-free original image is estimated by utilizing the available quantized data, knowledge about the quantizer stride size, and prior noesis about the smoothness of the original image.

A deterministic formulation of the problem is followed in [27]. Two solutions are developed for the removal of blocking artifacts in however images. The get-go one is based on the CLS formulation and a successive approximations iteration is utilized for obtaining the solution. The second approach is based on the theory of projections onto convex sets (POCS), which has found applications in a number of recovery issues. The evidence analysis inside the hierarchical Bayesian paradigm, mentioned to a higher place, is applied to the same trouble in [28]. Expressions for the iterative evaluation of the unknown parameters and the reconstructed epitome are derived. The human relationship betwixt the CLS-iteration adaptive successive approximations solution and the hierarchical Bayesian solution discussed in the previous section is besides applicative here.

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780123744579000159

Iterative Prototype Restoration

Aggelos K. Katsaggelos , Chun-Jen Tsai , in Handbook of Prototype and Video Processing (Second Edition), 2005

6.3 Additional Applications

In this affiliate we have concentrated on the application of the successive approximations iteration to the image restoration problem. Yet, as already mentioned multiple times already a number of recovery problems can notice solutions with the utilize of a successive approximations iteration. Two recovery problems which take been actively pursued in the final 10–15 years due to their theoretical claiming simply also their commercial significance, are the removal of compression artifacts and resolution enhancement.

The problem of removing pinch artifacts addresses the recovery of information lost due to the quantization of parameters during compression. More than specifically, in the majority of existing image and video compression algorithms the image (or frame in an paradigm sequence) is divided into square blocks which are processed independently from each other. The discrete cosine transform (DCT) of such blocks (representing either the image intensity when dealing with still images or intra-coding of video blocks or frames, or the displaced frame difference when dealing with inter-coding of video blocks or frames) is taken and the resulting DCT coefficients are quantized. As a consequence of this processing annoying blocking artifacts result, primarily at high compression ratios. A number of techniques have been developed for removing such blocking artifacts for both still images and video. For example, in [ 18, 27] the problem of removing the blocking artifacts is formulated as a recovery trouble, according to which an judge of the blocking artifact-complimentary original paradigm is estimated by utilizing the available quantized data, noesis near the quantizer step size, and prior cognition about the smoothness of the original image.

A deterministic formulation of the trouble is followed in [27]. Two solutions are developed for the removal of blocking artifacts in still images. The offset one is based on the CLS formulation and a successive approximations iteration is utilized for obtaining the solution. The second arroyo is based on the theory of projections onto convex sets (POCS), which has found applications in a number of recovery issues. The evidence analysis within the hierarchical Bayesian image, mentioned in a higher place, is applied to the aforementioned problem in [18]. Expressions for the iterative evaluation of the unknown parameters and the reconstructed epitome are derived. The relationship betwixt the CLS-iteration adaptive successive approximations solution and the hierarchical Bayesian solution discussed in the previous section is too applicable here.

Resolution enhancement (also referred to as super-resolution) is a problem which has besides seen considerable activity recently (for a recent review see [half dozen, xiii] and references therein). It addresses the problem of increasing the resolution of a single image utilizing multiple aliased depression-resolution images of the aforementioned scene with sub-pixel shifts among them. It also addresses the problem of increasing the resolution of a video frame (and consequently the whole sequence) of a dynamic video sequence past utilizing a number of neighboring frames. In this case the shifts between whatever two frames are expressed by the move field. The depression resolution images and frames can exist noisy and blurred (due to the image acquisition organization), or compressed, which further complicates the problem. There are a number of potential applications of this technology. It can be utilized to increment the resolution of any musical instrument by creating a number of images of the same scene, but besides to supervene upon an expensive loftier resolution instrument by one or more low resolution ones, or it can serve as a compression machinery. Some of the techniques developed in the literature accost, in addition to the resolution enhancement problem, the simultaneous removal of blurring and compression artifacts, i.e., they combine the objectives of multiple application mentioned in this affiliate. For illustration purposes consider the example shown in Fig. ten [xx]. In Fig. 10a the original high resolution image is shown. This prototype is blurred with a 4×4 non-causal compatible blur part and downsampled past a factor of 4 in each direction to generate 64 low resolution images with global sub-pixel shifts which are integer multiples of 1/4 in each management. Racket of the same variance was added to all depression resolution images (resulting in SNR of 30 dB for this example). 1 of the low resolution images (the one with zero shifts in each direction) is shown in Fig. 10b. The best bilinearly interpolated image is shown in Fig. 10c. A hierarchical Bayesian approach is utilized in [20] in deriving iterative algorithms for estimating the unknown parameters (image model parameter similar to Due south in Eq. (53) and the additive noise variance) and the loftier resolution image past utilizing the 16 low resolution images, assuming that the shifts and the blur are known. The resulting high resolution image is shown in Fig. 10d. The same experiment was performed merely with a dissimilar corporeality of noise added to each depression resolution image from a ready of 10 dB, twenty dB, or 30 dB SNR. The all-time bilinearly interpolated image is shown in this case in Fig. 11a while the one resulting from the hierarchical Bayesian approach in Fig. 11b. Finally, the same terminal experiment was repeated with the resolution chart image. One of the 16 low resolution images is shown in Fig. 12a, while the one generated by the hierarchical Bayesian arroyo is shown in Fig. 12b. The hierachical Bayesian approach was besides used for the recovery of a high resolution sequence from depression resolution and compressed observations [24].

Figure ten. Resolution enhancement case: (a) original image; (b) one of the 16 low resolution images with SNR = 30 dB (all 16 images have the same SNR); (c) best bilinearly interpolated image; (d) estimated high resolution epitome by the hierarchical Bayesian approach [twenty].

FIGURE 11. Resolution enhancement example: (a) best bilineraly interpolated low resolution image (the SNR for the low resolution images is at random either 10 dB or 20 dB or thirty dB); (b) estimated high resolution image by the hierarchical Bayesian arroyo (sixteen noise parameters) [20].

Figure 12. Resolution enhancement example: (a) i of the 16 low resolution images (the SNR for the depression resolution images is at random either 10 dB or 20 dB or 30 dB); (b) estimated high resolution image past the hierarchical Bayesian approach [20].

Read full affiliate

URL:

https://www.sciencedirect.com/scientific discipline/article/pii/B9780121197926500784

Warping-Based Motion Interpretation Techniques

Mohammed Ebrahim Al-Mualla , ... David R. Bull , in Video Coding for Mobile Communications, 2002

v.iv Discussion

Block matching methods have ever been criticized because of their simple uniform translational model. The argument against this model is that, in practice, a block tin can contain multiple moving objects and the motion is commonly more complex than simple translation. The shortcomings of this model may appear as poor prediction quality for objects with nontranslational movement and besides every bit blocking artefacts within move-compensated frames. Warping-based methods employing higher-club movement models have been proposed in the literature as alternatives to cake-matching methods. This affiliate investigated the performance of warping-based methods and compared it to that of block-matching methods. The results of this comparison have shown that despite their improvements over bones block-matching methods, the utilize of warping-based methods in applications similar mobile video communication may non be justifiable, due to the huge increment in computational complexity. In fact, similar (if not improve) improvements tin be obtained, at a fraction of the complexity, by just augmenting basic block-matching methods with advanced techniques like subpel accuracy and overlapped motion compensation. Ane can debate that warping-based methods can besides benefit from subpel accuracy and overlapped move compensation, as shown in Refs. 113 and 117, simply over again this volition further increase complexity. In improver to their loftier computational complexity, warping-based methods tin suffer from warping artefacts, poor compensation of covered/uncovered background, and lack of motion field segmentation. Reducing the complexity of warping-based methods and including them in a hybrid WBA/BMA video codec are two possible areas of further research.

Read full chapter

URL:

https://www.sciencedirect.com/scientific discipline/article/pii/B978012053079350007X