MPEG Video Compression Standards Quiz

2D Discrete Wavelet Transform (DWT)

• Used for image input of size NxN
• Convolve each row with h₀[n] and h₁[n], discard odd-numbered columns, concatenate
• Convolve each column of the result with h₀[n] and h₁[n], discard odd-numbered rows, concatenate
• One stage completes, resulting in four subbands (LL, HL, LH, HH)
• LL subband can be further decomposed for more decomposition levels

JPEG Standard

• Developed by the Joint Photographic Experts Group (JPEG)
• Formally accepted as an international standard in 1992
• A lossy image compression method
• Employs the Discrete Cosine Transform (DCT)
• 2D DCT converts image from spatial domain (f(i, j)) to frequency domain (F(u, v))

JPEG Observations

• Observation 1: Image intensity changes slowly over small areas (spatial redundancy)
• Observation 2: Humans are less sensitive to loss of high frequency components than low
• Observation 3: Visual acuity is higher for grayscale than color (4:2:0 chroma subsampling used)

JPEG Encoder Block Diagram

• Input: YIQ or YUV image
• DCT (8x8 blocks)
• Quantization (Q(u,v) matrix)
• Coding tables
• Entropy coding
• DPCM (DC coefficient)
• Zig-zag ordering
• Run-length encoding
• AC coefficients
• Header tables
• Data

DCT on Image Blocks

• Divides images into 8x8 blocks
• Appied 2D DCT to each block f(i, j)
• Generates DCT coefficients F(u, v) for each block
• Isolating each block from neighbouring context leads to "blocky" appearance at high compression

Quantization

• F(u, v) = round(F(u, v) / Q(u, v))
• Represents quantized DCT coefficients for JPEG entropy coding
• Main source of loss in JPEG compression
• Compression ratio can be changed multiplicatively scaling Q(u, v) matrix
• Quality factor in JPEG implementations is linearly tied to scaling factor

Quantization Tables

• Q(u, v) values tend to be larger towards lower right corner (higher loss at higher frequencies)
• Psychophysical studies determine default Q(u,v) values to maximize compression ratio whilst minimizing perceptual losses

Run-Length Coding (RLC) on AC Coefficients

• Converts F(u,v) values into sets { #-zero-to-skip, next non-zero value }
• Zig-zag scan turns 8x8 matrix into 64-vector to maximize zero runs

DPCM on DC Coefficients

• Differential Pulse Code Modulation (DPCM) for DC coefficients' coding
• Different from AC coefficient coding
• Computes differences between DC coefficients in successive image blocks

JPEG Modes

• Sequential Mode: Default mode, left-to-right, top-to-bottom scan
• Progressive Mode
• Hierarchical Mode
• Lossless Mode: Discussed in Chapter 7, replaced by JPEG-LS

JPEG 2000 Standard

• More successful and popular image format than JPEG
• Better rate-distortion tradeoff and improved subjective image quality
• Additional functionalities lacking in current JPEG standard
• Uses Embedded Block Coding with Optimized Truncation (EBCOT) algorithm
• Partitions wavelet transform subbands into code blocks
• Generates scalable bitstream for each code block, improving error resilience

Layer Formation and Representation

• JPEG 2000 offers resolution and quality scalability
• Two-tiered coding strategy using layered bitstream organization
• First tier produces embedded block bit-streams
• Second tier encodes block contributions to each quality layer

Region of Interest Coding in JPEG 2000

• Images may contain important information in certain regions (ROI)
• MAXSHIFT method scales ROI coefficients to higher bit-planes
• ROI decoded and refined before rest of image at reduced bit-rate

Problems with JPEG 2000

• Higher computational demands
• Higher memory demands

H.261

• Developed in 1990
• Video compression standard (MC based)
• Designed for videophone, video conferencing and audiovisual services over ISDN
• Bit-rates of p x 64 kbps and delayed less than 150ms

H.261 Frame Sequence

• I-frames: Independent images using JPEG-style transform coding
• P-frames: Dependent on previous P/I frame for prediction (forward predictive coding)
• Temporal redundancy included in P-frame, spatial in I-frames

Rate Control: Problem

• H.261 encodes varying bit rates over constant bit rate channels (e.g 384 kbps)

Rate Control: Solution

• Buffered encoded bitstream
• Increased quantization step size when buffer full
• Reduced quantization step size when buffer empty

H.263

• Improved video coding standard (ITU-T)
• Designed for low bit-rate communications (below 64 kbps)
• For H.324, H.323, RTP/IP, RTSP solutions, streaming media and SIP

Half-Pixel Precision

• H.263 supports half-pixel precision for motion vectors, to reduce prediction error.
• Uses bilinear interpolation to compute required half-pel pixel values

Optional H.263 Coding Modes

• Unrestricted motion vector mode: Motion vectors don't have to be within image boundary.
• Syntax-based Arithmetic Coding Mode: Uses VLC/variable length coding for DCT coefficients
• Advanced prediction mode: Four motion vectors from neighbouring blocks (left, right, above, below)
• PB frames mode: Uses B-frames, similar to MPEG prediction, bidirectionally from previous and future frames

MPEG Overview

• Established in 1988 for digital video development
• Defines only compressed bitstream to avoid proprietary interests
• Compression algorithms are manufacturer's responsibility

MPEG-1

• Approved in 1991, for moving picture/audio storage
• 1.5 Mbps bit-rates
• Common storage media: CD, VCDs.
• Uses specified CCIR601 digital TV format (SIF)
• Non-interlaced coding, at 30/25fps, uses 4:2:0 chroma subsampling

Motion Compensation in MPEG-1

• Based on MC
• Motion Estimation (ME) identifies best matching macroblock (MB) from previous I or P frame
• Prediction error between current & matching MB sent for DCT
• Uses forward prediction, previous frame as reference

Bidirectional Search

• Previous and next frames are potentially used for prediction

Motion Compensation (MC) B-frame Coding

• B-frames are predicted bi-directionally from previous and future frames, improving prediction accuracy
• Weighted average of matches from forward and backward predictions are used as reference

MPEG Frame Sequence

• I/P/B frame arrangement (display vs coding order)

MPEG-1 Video Bitstream

• Layers of sequence header,GOP headers, picture headers, macroblocks and blocks
• VLC encoding, run-length and Differential DC coefficient coding

MPEG-2

• Developed in 1994 & standard for higher quality video
• Uses bitrates above 4 Mbps, initially for digital broadcast
• Supports DVDs, digital video discs
• Defines seven coding profiles for applications (low delay, scalable, HDTV)
• Includes up to four levels per profile, defines display resolutions

MPEG-2 Scalabilities

• Provides flexibility via scalability in multiple domains:
• SNR: Improves quality
• Spatial: Higher resolution
• Temporal: Higher frame rates
• Hybrid: Combining features
• Data Partitioning: Distributing DCT high/low frequencies

SNR Scalability

• Base layer has coarse quantization for fewer bits, low-quality video
• Enhancement layer compares with original video and quantizes difference for coded refinement; bit-stream called Bits_enhance

Spatial Scalability

• Base layer produces lower resolution images; enhancement layer combines with base-layer predicted MB to get full resolution.

Temporal Scalability

• Input demultiplexed into two parts of reduced frame-rate, for encoding
• Base layer uses normal single-layer procedures to generate Bits_base

Hybrid Scalability

• Combining any of above three types of scalabilities

Data Partitioning

• Base partition contains lower frequencies in DCT coefficients, and enhancement higher
• Not considered layered coding; just breaks video data into separable partitions
• Useful for noisy channels and progressive transmission

Other Major Differences from MPEG-1

• Better bit error resilience (Transport Stream)
• 4:2:2 and 4:4:4 chroma for color quality
• Non-linear quantization structure
• More restricted slice structure
• More flexible video formats

MPEG-4 Overview

• Focuses on user interactivity beyond basic compression
• Uses object-based encoding
• Allows for composition, manipulation, and retrieval of visual objects
• Wide bit-rate range (5 kbps to 10 Mbps)

MPEG-4 Object Types, Profiles, and Levels

• Standardized profiles & levels ensure interoperability between implementations
• Specifies visual, audio, graphics, scene description and object descriptor profiles

MPEG-4 Part 10/H.264

• New video compression standard (from JVT), which is also called H.264
• Offers significant improvements (up to 30–50% better compression) vs. MPEG-2 or other standards
• Designed for HDTV content

MPEG-7

• Facilitates audiovisual content retrieval using metadata
• XML-based metadata is separate from actual encoding/storage

MPEG-21

• Defines a multimedia framework for multimedia distribution and consumption
• Introduces a Digital Item as fundamental unit for distribution and transaction
• Aims to provide technology for efficient item exchange/use

Test your knowledge on MPEG video compression standards, focusing on MPEG-1 and MPEG-2, their profiles, and applications. This quiz covers motion vectors, resolutions, and the principles of video encoding and compression. Challenge yourself with advanced questions about video coding features and standards.