[attachment=11885]
Spatial &Temporal Compression of Streaming Video for Digital TV Broadcast
Digital TV has come of age in India. It affords extremely good picture quality and stunning theatre-like sound effects. Besides, it offers a tremendous amount of user interactivity, something which is not possible in TVs using analog technologies such as NTSC/ PAL/ SECAM. With interactive TV, it is possible to perform on-line trading, conduct secure bank transactions, surf the internet, request viewing of selected programs, and perform airline/ train enquiries etc. The possibilities are endless. The interactivity can be achieved through Set-Top Boxes (Integrated Receiver – Decoder) interfaced to the conventional analog TV or on Desktop PCs under Multimedia Home Platform.
Digital TV technology uses digital bit-streams of audio, video and data multiplexed as a single stream using the MPEG2/ MPEG7 protocol. The transmitter digitizes the audio, video and data and generates the transport multiplex. Special modulation techniques such as QAM (Quadrature Amplitude Modulation) are used to convert the digital stream to RF signals which are aired through transmission towers. On the receiver side, the RF signal captured by the receiver’s antenna converts it into a digital stream. This digital stream is de-multiplexed to recover the audio, video and data. Extensive processing is done on the digital audio and video before these are played. Digital technology ensures impeccable quality and almost eliminates noise in the transport stream. Because digital technology employs software for signal processing, a great amount of flexibility is ensured and interactivity is easily achieved through a back-channel.
The heart of a Digital TV is the Codec (Compressor & De-Compressor) which performs the compression and de-compression of the audio and video samples on the fly. Various display resolution modes are supported by the Digital TV Standards and these are covered by ETSI Standards. As the resolution and the display quality increases, the memory requirements in storing pictures become more and more. For a Standard Definition TV (720x576), the memory requirement is 1.215 MB per frame. Assuming a scan rate of 30 frames/sec, the memory requirement for storing a 1 second worth of video recording is 36.45 MB. Even with computers of massive resources, such memory requirements are extremely difficult to meet. Therefore, there is a genuine requirement for compression of audio and video signals before generating the transport multiplex.
The theme of our project is concerned with the video part of the Digital TV transmission in building a powerful Codec – with features beyond those defined by the MPEG2/ MPEG7 and H264 Standards. Compression of progressive frames is achieved through transform coding techniques such as DCT/ Wavelets followed by minimum entropy coding. Two types of redundancies are exploited in the compression – (1) inter-frame redundancy and (2) temporal redundancy. Temporal redundancy can be brought out by estimation of motion vectors that can perform prediction of future frames. Temporal compression is therefore achieved by storing a reference frame along with a set of motion vectors. The standards for MPEG2 and H264 define broad guidelines for video compression but still recommend DCT for transform coding. In our project, we will design our own wavelets and implement them for transform coding. Further, we propose to use a new and hitherto unused technique based on heuristics and Neural Networks to perform motion estimation. The proposed algorithms are expected to yield 15 – 20% more compression, significant improvement in the quality of reconstructed frames and extremely large values for SNR, indicating less loss of information in the compression – decompression cycles.
The proposed technologies are:
• Custom-built Wavelets with very high compactness and energy retention
• New Heuristic and ANN Techniques in motion vector estimation
• New compression technique using Quad-Trees
The input to the system will be Transport Stream Files containing audio and video in our format. The entire system is proposed to be implemented in a combined environment of MATLAB 2006A and Visual C++.
Contemplated Extensions:
We propose to extend the functionalities by adding the following features:
1. Interactive Content
2. Feature Tracking
3. Replacement of Bill-Boards in Real Time
4. Elimination of Logos and Sub-Titles
Existing System versus Proposed System
Existing implementations of Digital TV in Europe and the United States employ Video Codec based on MPEG2 technology. MPEG2 employs DCT based compression. While 2D DCT is good in concentrating the energy of image in a few transform coefficients, better transform techniques based on Wavelets exist. While DCT is based on cosine waves of infinite duration, Wavelets are based on waves of finite duration with excellent localization and better energy concentration features. Therefore, any transform coding based on wavelets will produce better results – more compression for a given quality and extremely high SNR values. The reconstructed image will have fewer artifacts. We, therefore depart from the existing systems in employing wavelets as the transform of choice. In order to perform temporal compression, we use the upcoming technology based on Artificial Neural Networks with a fuzzy layer. The compression itself will be performed by a variation of the Quad Tree algorithm used by SPIHT (Set Partitioning in Hierarchical Trees) as opposed to the Huffman Coding used in MPEG2. With these technologies in place, we expect the following significant improvements as compared to MPEG2 technology:
1. Enhancement of compression ratio by 10 to 15%
2. Faster compression and decompression cycles
3. Improved Reconstructed Frames with larger SNR
4. Provision for interactivity while implemented on STBs
5. Provision for support of various video formats
These improvements are significant enough to qualify our implementation for practical use in commercial STBs. The entire system is proposed to be run under H264 test suites provided ETSI for certification towards commercial use.
Practical implementations of Digital TV employ forward error correction algorithms in order to correct errors which are introduced by the environment. While there are many approaches, Reed-Solomon algorithm fares very well in error recovery except under very abnormal circumstances. All Digital TV implementations (DVB, ATSC & ISDB) employ forward error correction techniques and we will include this feature also in building the Transport Multiplex.
Implementation Roadmap
The project is proposed to be implemented in the following stages:
 Justification for lossy compression of streaming video
 Study of MPEG2 Transport Streams and protocol
• MPEG2 Protocol – Transport Multiplex
• Inter Frame Compression
• Temporal Redundancy Compression
• Transform Coding based on 8x8 Discrete Cosine Transform
 Custom Wavelet based compression
 Reconstructed Video Quality Assessment
 Compression Percentage and Bit-Rate Issues
 Design of H264 Codec with enhanced features
 Generation of Transport Streams for Testing
 Testing of the Codec in Real Time conditions
 Performance Evaluation & Comparison with existing codecs
Run certification procedures with Test Suites

who wrote this topic