Document Sample

A DCT based Algorithm for Blocking Artifact Reduction from DCT Coded Images Vinay Kumar Srivastava Department of Electronics and Communication Engineering, Motilal Nehru National Institute of Technology, Allahabad, India E-mail: vksrivastaval2@rediffmail.com Abstract- The edge component causes the high frequency local statistics of the image. In these algorithms, the components in the transformed domain. Thus, examining the classification of image areas is used to capture the non- discrete cosine transform (DCT) coefficients of the block itself can identify the monotone blocks. Moreover, the artificial discontinuities in monotone area due to blocking artifacts also stationary behavior of the image. Classification Si based on available edge information extracted from the received blocky cause the high frequency components in DCT domain. Thus, these image. Hence the performance of space-variant or adaptive discontinuities are reflected as high frequency components in the filtering scheme degrades. Furthermore, the need of strong DCT of concatenated block of two or more neighboring monotone filter in monotone area and weak directional filter in edge area blocks. It is also found that there exists a relationship between the DCT coefficient of two monotone blocks and that of concatenated makes it complex In references [7]-[15], the blockig artfacts block. The DCT coefficients due to blocking artifacts are present are reduced by processing the image in DCT domain itself. The in the concatenated block but not in individual block. Therefore, DCT domain algorithms are efficient since the signal need not from the relationship between the DCT coefficients of monotone be compressed or decompressed. Recovery of accuracy loss in block of different sizes we can detect the high frequency DCT coefficient by discontinuity criterion is another method components due to blocking artifacts and thus we can eliminate that can reduce these artifacts efficiently in monotone area. them to reduce the blocking artifacts. Jeon and Jeong [8] proposed a method for post-processing that Keywords: Post-processing, Blocking artifacts, DCT, DCT-domain gives minimum discontinuity of pixel values over block processing, Block transform coding. boundaries by compensating the loss of a coefficient's I. INTRODUCTION accuracy in the transform domain. In [9], Hsia et al. proposed In block discrete cosine transform (BDCT) based image transform-domain algorithm to effectively classify the compression the blocking artifacts are main cause of characteristics of blocks and estimate the strength of the blocky degradation, especially at higher compression ratio. This is the effect. An adaptive fmite impulse response filter to effectively remove the blocky effect also proposed. In [10], Peak et al. main drawback of BDCT based image compression schemes and thus wavelet based image compression is used in new proposed a method in which two adjacent homogeneous blocks -.. . standard JPEG 2000. Most of the image compression standards .. ~~~from block boundar are found and the local frequency ry qu y (e.g., JPEG, MPEG, and H.263) use BDCT in their image characteristics in the homogeneous block are examined compression scheme due to its excellent energy compaction through DCT. Then, relation between the DCT coefficients of and de-correlation properties. DCT is used to exploit the spatial two homogeneous blocks of different sizes is derived. By correlation for image compression. In BDCT based image considering this information the high-frequency components compression, the image degradation becomes visible when mainly caused by the blocking artifact are detected. 'psoriexceeds certain level. hs erdhn compression ratio ecescranlvl These degradations Recently, Chen et al. [11] introduced a DCT-domain post- manifest themselves as blocking artifacts due to the rigid block filtering approach to reduce blocking artifacts, where the post- partitioning of the image and ringing noise around edges due to filter made use of the DCT coefficients of shifted blocks in coarse quantization. Both effects are visually annoying and order to obtain a close correlation between the DCT have substantial impact on the subjective quality of image. At coefficients at the same frequency. In [12], Chang and higher compression ratios, since very few coefficients are Messerschmitt, proposed algorithms to manipulate compressed encoded and each coefficient must be represented at a very video in the transform domain. In [13], a method for efficiently assessing, and subsequently reducing, the severity of blocking visible, visible. artifacts in compressed image bitstreams is proposed. In the Linear, space-invariant filtering is inadequate to remove algorithm, blocking artifacts are modeled as 2-D step these artifacts whereas iterative methods such as projections functions. A fast DCT-domain algorithm extracts all onto convex sets (POCS) have greater computational parameters needed to detect and estimate the blocking artifacts, c post-processingpxySellrh [1]- av e by exploiting several blockiness, ofnovelhuman vision method Using the estimate of properties a the DCT-domain system. been proposed to reduce blocking artifacts Of blocki coding 1 system. Out of these algorithms space-variant / adaptive is then developed which adaptively reduces detected blocking filtering techniques are more attractive as they try to exploit the ariacs Lu an Wad[4 rpsdatcnqe hc preserved the edge and texture information. This adaptive 1-4244-0726-5/06/$20.OO '2006 IEEE 2815 Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply. approach performs blocking artifact reduction in both the DCT The DCT of concatenated row can be expressed as: and spatial domains. For smooth regions, the continuity of original pixel levels in the same block and the correlation 1 k /2 between the neighboring blocks is used to reduce the 2 discontinuity of the pixels across the boundaries. For texture and edge regions an edge-preserving smoothing filter is for k=0,2,4, applied. Zhao et al. [15], proposed a DCT domain deblocking technique by shifting the blocks successively in horizontal and F 1 vertical directions, respectively, and simultaneously shrinking W(k) )cos7(2n + 1)k c the undesired high frequency DCT coefficients. a(k) N -L 4 + This paper is organized as follows. In Section II, the 1) _ algorithm to reduce blocking artifacts is described. Simulation 2 /1) v(n)sin ;T(2n + I)k results are presented in Section III. Finally, conclusions are L 4N given in Section IV. 11. PROPOSED SCHEME for k = 1,3,5,... (4) Blocking artifacts are mainly due to independent coding of From the above equation (4) it is clear that the even numbered (2kth) DCT coefficient of W depends only on the kth different blocks. The edge components of image cause the high frequency components in the transformed domain. Thus, DCT of Uand Vwhereas the odd numbered DCT coefficients examining the DCT coefficients of the block itself can identify coefficients of W are expressed as a weighted sum of u(n) and the monotone or edge block. The artificial discontinuities in v(n). Thus, only the odd numbered DCT coefficients of W are monotone area are due to blocking artifacts. These affected by the artificial discontinuities in the block. Assuming discontinuities also cause the high frequency components. In that the original image is highly correlated and the global the proposed algorithm two horizontally adjacent blocks of size frequency characteristics in two adjacent blocks should be N x N are concatenated to make a new block of size 2N x N. similar to the local frequency characteristics in each block. The discontinuities due to blocking artifacts are reflected as Thus, the high-frequency components in the global high frequency components in the DCT of concatenated block characteristics of a decoded image, which are not found in the of two or more neighboring monotone blocks. It is also found local ones, can be considered as a result from the blocking that there exists a relationship between the DCT coefficient of artifact. Here, N-point DCT will be employed to obtain the two monotone blocks and that of concatenated block. The DCT local characteristics, and 2N-point DCT to obtain the global coefficients due to blocking artifacts are present in the ones. The relation between N-point and 2N-point DCT concatenated block but not in individual block. Therefore, from coefficients is used to detect the undesired high frequency the relationship between the DCT coefficients of monotone components, mainly caused by the blocking artifact. If the block of different sizes we can detect the high frequency original image is highly correlated so that the block components due to blocking artifacts and thus we can eliminate discontinuities are invisible, then the odd-numbered DCT them to reduce the blocking artifacts. The algorithm is also coefficients of W of the original image can be approximated by applied to vertically adjacent blocks to reduce the artificial interpolating adjacent even-numbered coefficients. Thus, from discontinuities due to blocking artifacts in horizontal direction. equation (4), the nonzero DCT coefficients of W are kept To find the relationship between N x N DCT and 2N x N within the range two times larger than those of U and V. In the DCT we extend the results obtained in [10]. The two adjacent decoded image, if nonzero coefficients of W occur at higher rows u(n) and v(n) are concatenated as w(n) and their DCT are locations than twice the highest nonzero valued location of U given by U(k), V(k) and W(k), respectively, as and V then those coefficients are believed to be the results of N-1 Fz(2n + l)k 1artificial discontinuities due to the blocking artifact. Thus, U(k) = a(k) L u(n) cos (1) actual nonzero coefficients due to signal are kept within the n=O _ 2N _J range two times larger than those of U and V. For concatenated 2D blocks, actual nonzero DCT V(k) = r(k) L v(n) cosl 1 (2) coefficients are present within the range two times larger than n=O L 2N j those of DCT coefficients of constituent block for all rows and column. The coefficients beyond this range are the coefficients 1 2N-1 FT(2n + 1)k due to the blocking artifacts. In proposed algorithm, the DCT W(k) a(k) w(n)cos , (3) coefficients in the first row and first column that are beyond this range are eliminated. Only first row and first column are 1 2 selected because of the effect of vertical and horizontal where oc(k) =|for k=0 and c~(k) =| for other values. discontinuities which has greater values in these locations. This N N algorithm has low computational complexity, as no filterinLg is 2816 Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply. required. The availability of fast algorithm for DCT corresponding index of concatenated block will increase and computation makes this algorithm suitable for low bit rate the edges will not be blurred by elimination. coding applications. The selection of first row and first column A. Corner Outliers Detection and Replacement for elimination further reduces the computation time. As discussed in [4], corner outliers are visible at the cross Following are the main steps of proposed Block Boundary of MxM blo Dicntnite Reuto Aloitm point oor Mblck when the corner pixel of one block iS very large very small as compared to corner pixels of the Step 1: Identification of block as monotone block: Actual neighboring blocks. The corner outliers detection and edges are blurred due to elimination of high frequency replacement algorithm [4] is used here. components. To protect the actual edges, this algorithm requires the identification of monotone blocks. In proposed algorithm, we selected only first row and first column for A3 A2 B2 B3 elimination of some DCT coefficients. Therefore, this step can Al A B B, be bypassed. In fact, in implementation of this algorithm, this - Block -C-D-D<I step is ignored because this algorithm doesn't produce much Cl C D boundary undesirable blur as discussed above. C3 C2 D2 D3 Step 2: Joining two horizontally adjacent monotone blocks and transforming the concatenated block in DCT In this algorithm, each corner pixel is compared with its domain: The two horizontally adjacent monotone blocks of neighbors to detect the corner outliers whereas to reduce this size M x M are concatenated into one block of size 2M x M. effect the corner A and its neighbors (A1 and A2) are replaced The DCT coefficients due blocking artifacts are present in the by its weighted average as given below: concatenated block but not in individual block. The 2M x M DCT of concatenated block gives the global characteristics. A = (4A + B + C + 2D + 4)/8 The value of M is taken as 8 for this simulation. A - (A + 3A + 2)/4 (5) Step 3: Finding the maximum non-zero valued index for A2 = (A' + 3A2 + 2) / 4 the concatenated block for elimination of DCT coefficients corresponding to blocking artifacts: The global frequency III. RESULTS characteristics in two adjacent blocks are similar to local ones in each block. This concept is used in identifying and DIn this section results of the proposed Block Boundary elimiatin th coffiiet du to blckn artiat the th eliminating in Discontinuities Reduction cefcetdLenna image are presented. algorithm applied on 256 x with The algorithm is implemented 256 concatenated block. For two blocks, the index of first row and first column are found by checking them for non-zero values M\ATLAB. To get blocky . image. at. different quality factors,'.a . . greater than a threshold. The practical value of threshold is block imaGgorocess blocky image S post-processed by the proposedalgorithm. proposed algorithm. found during simulation. The maximum non-zero valued index for any row or column of a block is defined as the location Various results are given in Table I and II Various quality beyond which all the DCT coefficients are greater than a measures are calculated to see the performances of the tresod For cocteae blck it is fon tha th thehod Fo cocteae blck it is fon tha the algorithms. The results of the algorithm are compared with baseline JPEG like Dmg decoded image. Peak signal to noise coefficients beyond the twice the maximum of these index basin JPEG likeDCT r values in respective dimension for two individual blocks, are zero. Thus, in concatenated block if the coefficient beyond this index exists then it is assumed that they are due to blocking PSNR In dB =M10 g10 r i2552 6 (6) SE artifacts and thus they are eliminated. In this algorithm, the selection of first row and first column for elimination further where 255 is the peak signal for 8 bit PCM. MSE is mean reduces the computation time. square error given by: MSE= 1I -2 Y a N-1 N-1 ~ \ Step 4: Repeat the above all steps for vertically adjacent (7) blocks: To reduce the artificial discontinuities due to blocking N2 L i=o j=0 j artifacts in the horizontal direction, all the steps of the algorithm are repeated by combining the two vertically where 4, and 4, are the pixel values at position (i, j) of adjacent blocks. original and decoded image respectively. Another reason for ignoring the first step is that the index of Subjective quality of an image depends on the properties of concatenated block is dependent on the maximum value of the human visual system (HVS), as discontinuities are more visible index in two dimensions for the two individual blocks, in monotone or slowly varying areas. Therefore, the PSNR is Therefore, if one or both blocks are edge block then the only the rough indicator of image quality and does not reflect the blocking artifacts. A new discontinuity measure is defined. 2817 Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply. - (a) (b) (c) (d) m m~ ~ ~ ~ ~ ~ ~ ~ ~ ~ (e) (f) Fig.1: Post-processed Lenna image compressed at 0.25386 bits/pixel by various algorithms. (a) DCT decoded. (b) Proposed Block Boundary Discontinuities Reduction Algorithm (c) With corner outliers reduction. (d), (e) and (f) gives corresponding results for Lenna image compressed at 0.30483. 2818 Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply. .. . .... .. ................... .... ....... .................... ................ ...................................................................................................................... ... ....... ...................... (a) (b) (c) Fig.2: Enlarged photographs of details of deblocked Lenna image compressed at 0.25386 bits/pixel by various algorithms. (a) DCT decoded. (b) Proposed Block Boundary Discontinuities Reduction Algorithm (c) With corner outliers reduction. .......................................... ...................................................... ................................. ..... ......... ..................... . .. ........................ ............................................................ ................................. ...... ........... ................................ Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply. TABLE I Although this algorithm look very simple but this algorithm is PERFORMANCE OF THE PROPOSED ALGORITHMS ON COMPRESSED LENNA g g ry ag able to reduce blockig artifacts significantly without much IMAGE: COMPARISON OF PSNR (IN dB) OBTAINED BY VARIOUS ALGORITHMS c s Post- processed by computational complexity. The performance of this algorithm Compression Block Boundary With Corner is better at high compression ratios; therefore it is suitable for DCT low bit rate videoin coding applications. The DCTartifacts in bits per pixel Decoded (bpp) Discontinuities Reduction Outliers Reduction algorithm proposed this paper reduces the blocking based Algorithm by eliminating some DCT coefficients of concatenated block of 0.30483 26.379 26.647 2.5739 two adjacent blocks. This algorithm has low computational 0.34958 27.338 27.550 27.621 complexity, as no filtering is required. The availability of fast 0.37165 27.701 27.915 27.979 algorithm for DCT computation makes this algorithm suitable 0.40969 28.337 28.510 28.537 for low bit rate coding applications. The selection of first row 0.44725 28.844 29.004 29.022 and first column for elimination further reduces the 0.48188 29.268 29.386 29.385 0.51541 29.636 29.737 29.720 computation time. The results show that this algorithm 0.54321 29.936 30.026 30.005 provides very good performance with minimum computational complexity. TABLE II REFERENCES PERFORMANCE OF THE PROPOSED ALGORITHMS ON COMPRESSED LENNA IMAGE: COMPARISON OF BPSNR (IN dB) OBTAINED BY VARIOUS ALGORITHMS [1] B. Ramamurthi and A. Gersho, "Non-linear space-variant post processing of block coded image", IEEE Trans. Acoustics, Speech and Signal Post-processed Processing, vol. ASSP-34, no. 5, pp. 1258-1268, Oct. 1986. Compression in by Block With Corner [2] C.J. Kuo and R.J. Hsieh, "Adaptive post-processor for block encoded Boundary images", IEEE Trans. Circuit Syst. Video Technol., vol. 5, no. 4, pp. 298- (bltsperp) bits per pixel (bPP) xe DCT Decoded DCTDeoded |Discontinuities Outliers Reduction Reduction Reduti|n304, [3] Aug. 1995. Y.L. Lee, H.C. Kim, and H.W. Park, "Blocking effect reduction of JPEG Algorithm images by signal adaptive filter ring", IEEE Trans. on Image Processing, 0.25386 15.930 16.523 16.773 vol. 7, no. 2, pp. 229-234, Feb. 1998. 0.30483 17.358 17.838 18.033 [4] H. W. Park and Y. L. Lee, "A post-processing method for reducing 0.34958 18.324 18.688 18.837 quantization effects in low bit-rate moving picture coding", IEEE Trans. 0.37165 18.748 19.125 19.259 Circuit Syst. Video Technol., vol. 9, no. 1, pp. 161-171, Feb. 1999. 0.40969 19.438 19.738 19.795 [5] S.D. Kim, J. Yi, H.M. Kim, and J.B. Ra, "A deblocking filter with two 0.44725 19.971 20.265 20.305 separate modes in block based video coding", IEEE Trans. Circuit Syst. 0.48188 20.528 20.747 20.746 Video Technol., vol. 9, no. 1, pp. 156-160, Feb. 1999. 0.51541 20.946 21.140 21.103 [6] V.K. Srivastava and G.C. Ray, "Design of 2D-multiple Notch Filter and 0.54321 21.246 21.407 21.361 Its Application in Reducing Blocking Artifact from DCT Coded Image", in Proc. IEEE Int. Conf. Engineering in Medicine and Biology, Chicago 2000, pp. 2829-2833, 23-28 July, 2000. The block boundary PSNR (BPSNR) is defined in the same [7] V.K. Srivastava, "Post-processing of DCT coded images", PhD way as PSNR but only one pixel from both side of block dissertation, IIT Kanpur, India, Jan. 2001. [8] B. Jeon and J. Jeong, "Blocking artifacts reduction in image compression boundary are considered for the calculation of MSE. with block boundary discontinuity criterion", IEEE Trans. Circuit Syst. The selective attenuation of AC component corresponding to Video Technol., vol. 8, no. 3, pp. 345-357, June 1998. block boundary discontinuities shows significant reduction in [9] S.C. Hsia, J.F. Yang, and B.D. Liu, "Efficient postprocessor for blocky blocking artifacts. Finally corner outliers detection and effect removal based on transform characteristics", IEEE Trans on circuits and systems for video Technol., vol. 7, no. 5, pp. 924-929, Dec. replacement algorithm is applied to further improve the 1997. performance. The performance of the algorithm improves as [10] H. Peak, R.C. Kim, and S.U. Lee, "A DCT-based spatially adaptive post- compression ratio increases. The reduction of discontinuity processing technique to reduce the blocking artifacts in transform coded images", IEEE Trans. Circuit System Video Technol., vol. 10, no. 1, pp. signifies that (see Table I and II) blocking artifacts are reduced 36-41, Feb. 2000. and it can also be observed visually in Fig. 1, 2 and 3. It can [11] T. Chen, H. R. Wu, and B. Qiu, "Adaptive postfiltering of transform also be seen from Table I and Table II that the improvement in coefficients for the reduction of blocking artifacts," IEEE Trans. Circuits Syst. Video Technol., vol. 11, pp. 594-602, May 2001. PSNR for block boundary pixels is more as compared to over [12] S.-F. Chang and D. G. Messerschmitt, "Manipulation and compositing of all PSNR. MC-DCT compressed video," IEEE J. Select. Areas Commun., vol. 13, Thus, proposed approach for discontinuity reduction reduces no. 1, pp. 1-11, Jan. 1995. [13] Shizhong Liu, and Alan C. Bovik, "Efficient DCT-Domain Blind the blocking artifacts eficintl. Ths cmpuatinaly efficient the locin rtiact efficiently. This computationally eficent Measurement and Reduction of Blocking Artifacts", IEEE trans. on algorithm also gives very good result in terms of PSNR, circuits and systems for video technol., vol. 12, no. 12, pp. 1139-1149, discontinuity measure (BPSNR) and the visual quality. It is December 2002. clear from the results that this algorithm gives very good [14] Y. Luo, R.K. Ward, "Removing Trans. Image Process., of block-based DCT compressed images", IEEE the blocking artifacts vol. 12, no. 7, performance with minimum computational complexity. pp. 838-842 July 2003. IV CONCL I* COCUSIN SIONS ~~~~~[15]Y. Zhao,reduction inand S. domain", Electronics Letters, vol.for blocking artifacts G. Cheng DCT Yu, "Post-processing technique 40, no. 19, In this paper, a simple yet effective algorithm for reducing p.29-3,1tSeemr204 blocking artifacts from DCT coded image is proposed. 2820 Authorized licensed use limited to: Aricent Technologies (Holdings) Ltd. Downloaded on February 9, 2010 at 02:49 from IEEE Xplore. Restrictions apply.

DOCUMENT INFO

Shared By:

Categories:

Tags:

Stats:

views: | 48 |

posted: | 6/9/2011 |

language: | English |

pages: | 6 |

OTHER DOCS BY rajat231

How are you planning on using Docstoc?
BUSINESS
PERSONAL

By registering with docstoc.com you agree to our
privacy policy and
terms of service, and to receive content and offer notifications.

Docstoc is the premier online destination to start and grow small businesses. It hosts the best quality and widest selection of professional documents (over 20 million) and resources including expert videos, articles and productivity tools to make every small business better.

Search or Browse for any specific document or resource you need for your business. Or explore our curated resources for Starting a Business, Growing a Business or for Professional Development.

Feel free to Contact Us with any questions you might have.