Basic Video Compression Techniques PDF

Chapter 10 Basic Video Compression Techniques  Outline 10.1 Introduction to Video Compression 10.2 Video Compression with Motion Compensation 10.3 Search for Motion Vectors 10.4 H.261 10.5 H.263 2 ...

Chapter 10 Basic Video Compression Techniques  Outline 10.1 Introduction to Video Compression 10.2 Video Compression with Motion Compensation 10.3 Search for Motion Vectors 10.4 H.261 10.5 H.263 2  10.1 Introduction to Video Compression  A video consists of a time-ordered sequence of frames, i.e., images.  An obvious solution to video compression would be predictive coding based on previous frames.  Compression proceeds by subtracting images: subtract in time order and code the residual error.  It can be done even better by searching for just the right parts of the image to subtract from the previous frame. 3  10.2 Video Compression with Motion Compensation  Consecutive frames in a video are similar - temporal redundancy exists.  Temporal redundancy is exploited so that not every frame of the video needs to be coded independently as a new image.  The difference between the current frame and other frame(s) in the sequence will be coded - small values and low entropy, good for compression. 4  Video Compression with Motion Compensation  Steps of Video compression based on Motion Compensation (MC): 1. Motion estimation (motion vector search). 2. MC-based Prediction. 3. Derivation of the prediction error, i.e., the difference. 5  Motion Compensation  Each image is divided into macroblocks of size N×N. By default, N = 16 for luminance images. For chrominance images, N = 8 if 4:2:0 chroma subsampling is adopted. 6  Motion Compensation  Motion compensation is performed at the macroblock level. The current image frame is referred to as Target Frame. A match is sought between the macroblock in the Target Frame and the most similar macroblock in previous and/or future frame(s) (Reference frame(s)). The displacement of the reference macroblock to the target macroblock is called a motion vector MV. 7 Fig. 10.1: Macroblocks and Motion Vector in Video Compression. 8  Figure 10.1 shows the case of forward prediction in which the Reference frame is taken to be a previous frame.  MV search is usually limited to a small immediate neighborhood – both horizontal and vertical displacements in the range [−p, p]: This makes a search window of size (2p+1)×(2p+1). 9  10.3 Search for Motion Vectors  The difference between two macroblocks can then be measured by their Mean Absolute Difference (MAD) 1 N- 1 N- 1 MAD (i , j ) = 2 å å C ( x + k , y + l ) - R( x + i + k , y + j + l ) N k =0 l =0 N : size of the macroblock, k and l : indices for pixels in the macroblock, i and j : horizontal and vertical displacements, C(x+k, y +l) : pixels in macroblock in Target frame, R(x+i+k, y +j +l) : pixels in macroblock in Reference frame. 10  Search for Motion Vectors  The goal of the search is to find a vector (i, j) as the motion vector MV = (u,v), such that MAD(i, j) is minimum: (u, v ) = [(i , j ) | MAD(i , j ) is minimum, i Î [- p, p], j Î [- p, p] ] 11  Sequential Search  Sequential search: sequentially search the whole (2p+1)×(2p+1) window in the reference frame (also referred to as full search or exhaustive search). A macroblock centered at each of the positions within the window is compared to the macroblock in the Target frame pixel by pixel and their respective MAD is then derived The vector (i, j) that offers the least MAD is designated as the MV (u, v) for the macroblock in the Target frame. 12  Sequential search method is very costly Assuming each pixel comparison requires three operations (subtraction, absolute value, addition), the cost for obtaining a motion vector for a single macroblock is 2 2 2 (2 p +1) ×(2 p +1) ×N ×3 Þ O ( p N ) 13 PROCEDURE 10.1 Motion-vector: sequential-search 14  2D Logarithmic Search  Logarithmic search: a cheaper version, that is suboptimal but still usually effective.  The procedure for 2D Logarithmic Search of motion vectors takes several iterations and is akin to a binary search: Initially only nine locations in the search window are used as seeds for a MAD-based search; they are marked as ‘1’. 15  After the one that yields the minimum MAD is located, the center of the new search region is moved to it and the step-size (offset) is reduced to half. In the next iteration, the nine new locations are marked as ‘2’, and so on. 16 Fig. 10.2: 2D Logarithmic Search for Motion Vectors. 17 PROCEDURE 10.2 Motion-vector: 2D-logarithmic-search 18  Using the same example as in the previous subsection, the total operations per second is dropped to: 19  Hierarchical Search  The search can benefit from a hierarchical (multiresolution) approach in which initial estimation of the motion vector can be obtained from images with a significantly reduced resolution.  Figure 10.3: a three-level hierarchical search in which the original image is at Level 0, images at Levels 1 and 2 are obtained by down-sampling from the previous levels by a factor of 2, and the initial search is conducted at Level 2.  Since the size of the macroblock is smaller and p can also be proportionally reduced, the number of operations required is greatly reduced. 20 Fig. 10.3: A Three-level Hierarchical Search for Motion Vectors. 21 Table 10.1 Comparison of Computational Cost of Motion Vector Search based on examples 22  10.4 H.261  H.261: An earlier digital video compression standard, its principle of MC-based compression is retained in all later video compression standards.  The standard was designed for videophone, video conferencing and other audiovisual services over ISDN.  The video codec supports bit-rates of p×64 kbps, where p ranges from 1 to 30.  Require that the delay of the video encoder be less than 150 msec so that the video can be used for real-time bidirectional video conferencing. 23 Table 10.2 Video Formats Supported by H.261 24 Fig. 10.4: H.261 Frame Sequence. 25  H.261 Frame Sequence  Two types of image frames are defined: Intra-frames (I-frames) and Inter-frames (P-frames):  I-frames are treated as independent images. Transform coding method similar to JPEG is applied within each I-frame.  P-frames are not independent: coded by a forward predictive coding method (prediction from previous I-frame or P-frame is allowed). 26  H.261 Frame Sequence  Temporal redundancy removal is included in P-frame coding, whereas I- frame coding performs only spatial redundancy removal.  To avoid propagation of coding errors, an I-frame is usually sent a couple of times in each second of the video.  Motion vectors in H.261 are always measured in units of full pixel and they have a limited range of ±15 pixels, i.e., p = 15. 27 Intra-frame (I-frame) Coding Fig. 10.5: I-frame Coding. 28  Intra-frame (I-frame) Coding  Macroblocks are of size 16×16 pixels for the Y frame, and 8×8 for Cb and Cr frames, since 4:2:0 chroma subsampling is employed.  A macroblock consists of four Y, one Cb, and one Cr 8×8 blocks.  For each 8×8 block a DCT transform is applied, the DCT coefficients then go through quantization, zigzag scan, and entropy coding. 29  Inter-frame (P-frame) Coding Fig. 10.6: H.261 P-frame Coding Based on Motion Compensation. 30  Inter-frame (P-frame) Coding  For each macroblock in the Target frame, a motion vector is allocated by one of the search methods discussed earlier.  After the prediction, a difference macroblock is derived to measure the prediction error.  Each of these 8x8 blocks go through DCT, quantization, zigzag scan and entropy coding procedures. 31  Inter-frame (P-frame) Coding  The P-frame coding encodes the difference macroblock (not the Target macroblock itself).  Sometimes, a good match cannot be found, i.e., the prediction error exceeds a certain acceptable level.  The MB itself is then encoded (treated as an Intra MB) and in this case it is termed a non-motion compensated MB.  For motion vector, the difference MVD is sent for entropy coding:  MVD = MVPreceding −MVCurrent 32  Quantization in H.261  The quantization in H.261 uses a constant step size, for all DCT coefficients within a macroblock.  If we use DCT and QDCT to denote the DCT coefficients before and after the quantization, then for DC coefficients in Intra mode: For all other coefficients: scale - an integer in the range of [1, 31]. 33  H.261 Encoder and Decoder  Fig. 10.7 shows a relatively complete picture of how the H.261 encoder and decoder work.  A scenario is used where frames I, P1, and P2 are encoded and then decoded.  Note: decoded frames (not the original frames) are used as reference frames in motion estimation.  The data that goes through the observation points indicated by the circled numbers are summarized in Tables 10.3 and 10.4. 34  I 0 I I I original image I decoded image Fig. 10.6(a): H.261 Encoder (I-frame). 35 I decoded image 0 I I Fig. 10.6(b): H.261 Decoder (I-frame). 36  P1 D1 D1 P1  P1' P1' D1 P 1  D1  P1 ' P1 P1' P1 original image P1' prediction D1 prediction error P1 decoded image D1 decoded prediction error Fig. 10.6(a): H.261 Encoder (P-frame). 37 P1' prediction P1' D1 P1 P1' D1 decoded prediction error P1 decoded (reconstructed) image Fig. 10.6(b): H.261 Decoder (P-frame). 38 39 P1' P1 I P1 Fig. 10.1: Macroblocks and Motion Vector in Video Compression. 40 P1 P1 D1 I P1' Fig. 10.6: H.261 P-frame Coding Based on Motion Compensation. 41  Syntax of H.261 Video Bitstream  Fig. 10.8 shows the syntax of H.261 video bitstream: a hierarchy of four layers: Picture, Group of Blocks (GOB), Macroblock, and Block. 1. The Picture layer: PSC (Picture Start Code) delineates boundaries between pictures. TR (Temporal Reference) provides a time-stamp for the picture. 42 2. The GOB layer: H.261 pictures are divided into regions of 11×3 macroblocks, each of which is called a Group of Blocks (GOB).  Fig. 10.9 depicts the arrangement of GOBs in a CIF or QCIF luminance image.  For instance, the CIF image has 2×6 GOBs, corresponding to its image resolution of 352×288 pixels. Each GOB has its Start Code (GBSC) and Group number (GN).  In case a network error causes a bit error or the loss of some bits, H.261 video can be recovered and resynchronized at the next identifiable GOB. 43 3. The Macroblock layer: Each Macroblock (MB) has its own Address indicating its position within the GOB, Quantizer (MQuant), and six 8×8 image blocks (4 Y, 1 Cb, 1 Cr). 4. The Block layer: For each 8x8 block, the bitstream starts with DC value, followed by pairs of length of zero-run (Run) and the subsequent non- zero value (Level) for ACs, and finally the End of Block (EOB) code. The range of Run is [0, 63]. Level reflects quantized values - its range is [−127; 127] and Level ≠ 0. 44 Fig. 10.8: Syntax of H.261 Video Bitstream. 45 Fig. 10.9: Arrangement of GOBs in H.261 Luminance Images. 46  10.5 H.263  H.263 is an improved video coding standard for video conferencing and other audiovisual services transmitted on Public Switched Telephone Networks (PSTN).  Aims at low bit-rate communications at bit-rates of less than 64 kbps.  Uses predictive coding for inter-frames to reduce temporal redundancy and transform coding for the remaining signal to reduce spatial redundancy (for both Intra-frames and inter-frame prediction). 47 Table 10.5 Video Formats Supported by H.263 48  H.263 & Group of Blocks (GOB)  As in H.261, H.263 standard also supports the notion of Group of Blocks (GOB).  The difference is that GOBs in H.263 do not have a fixed size, and they always start and end at the left and right borders of the picture.  As shown in Fig. 10.10, each QCIF luminance image consists of 9 GOBs and each GOB has 11×1 MBs (176×16 pixels), whereas each 4CIF luminance image consists of 18 GOBs and each GOB has 44×2 MBs (704×32 pixels). 49 Fig. 10.10: Arrangement of GOBs in H.263 Luminance Images. 50  Motion Compensation if H.263  The horizontal and vertical components of the MV are predicted from the median values of the horizontal and vertical components, respectively, of MV1, MV2, MV3 from the “previous", “above" and “above and right" MBs (see Fig. 10.11 (a)).  For the Macroblock with MV(u; v): 51 Fig. 10.11: Prediction of Motion Vector in H.263. 52  Half-Pixel Precision  In order to reduce the prediction error, half-pixel precision is supported in H.263 vs. full-pixel precision only in H.261.  The default range for both the horizontal and vertical components u and v of MV(u, v) are now [−16, 15.5].  The pixel values needed at half-pixel positions are generated by a simple bilinear interpolation method, as shown in Fig. 10.12. 53 Fig. 10.12: Half-pixel Prediction by Bilinear Interpolation in H.263. 54