Low Complexity AVS China Part II Video Codec Design using Data mining Techniques Comprehensive proposal May 6th 2010 by Jennie G Abraham Supervising Professor Dr K R Rao Outline Introduction AVS China Video Coding Standard Identify area of research Problem statement Proposed solution Summary 1 INTRODUCTION Fig 1 1 Example of a multimedia ecosystem Audio Video History Fig 1 2 History of audio video coding standards 2 AVS China Video Coding Standard AVS China Part 2 is the official video coding standard of China developed by the Audio Video Standard Working Group of China 1 Table 2 1 Different parts of AVS standard 4 AVS China Part II Profiles Profile subset of syntax semantics and algorithms defined by AVS Profiles Key Applications Jizhun baseline profile Television broadcasting HDTV etc Jiben basic profile Mobility applications etc Shenzhan extended profile Video surveillance etc Jiaqiang enhanced profile Multimedia entertainment etc Table 2 2 Application based profiles of AVS 11 Profile Feature Comparison Table 2 3 Features of the four profiles in AVS China video codec 11 AVS P2 Layered Structure Chrominance Cb Cr Macroblock Group of Picture s GOP Luminance Y Slice Pictu re 8 pixels Block Video input 8 pixels Fig 2 2 Layered structure of the input data UMCP ENEE408G Slides created by M Wu R Liu 2002 RGB vs YUV RGB YUV Luminance and Chrominance 4 4 4 format Y Cb Cr Input Video Formats Video Format AVS video codec input format Fig 2 1 Color formats of input video AVS Codec Architecture Figure 2 3 AVS video coding framework 7 Picture Types Three types of pictures are defined by AVS 1 Intra pictures I pictures 2 Predicted pictures P pictures 3 Interpolated pictures B pictures Prediction 1 Intra Prediction I frames 2 Inter Prediction Forward prediction P frames Bi directional prediction B frames Intra Prediction Predict Previously coded pixels Previously coded pixels Mode decisions Predict Current block 8 x 8 Figure 2 6 AVS video coding framework 7 Inter Prediction Figure 2 4 Inter Prediction forward and backward prediction 19 AVS China Macroblock Type AVS China P2 4 modes AVS China P7 all 8 modes Fig 2 5 Macroblock type for inter prediction 3 3 Mode Decision Computationally most expensive process in video coding Ripe area to investigate complexity reduction Focus area of the current research Constraints Maintain the quality Maintain the compression ratio Intra Mode Decisions Luminance Y Figure 3 1 Five luminance intra prediction modes 21 Mode 0 Vertical mode Mode mode Mode Mode Mode 1 Horizontal 2 DC mode 3 Down left 4 Down right Intra Mode Decisions Chrominance Figure 3 2 Four chrominance intra prediction modes 21 0 1 2 3 DC mode Horizontal mode Vertical mode Plane mode Example Intra Mode Decision Already decoded Being currently predicted Yet to be decoded Figure 3 3 Intra mode decision example 10 Intra Mode Decision Example cont d Mode 0 Mode 1 Mode 2 Vertical modeHorizontal mode DC mode SAE 657 SAE 619 SAE 607 Mode 3 Down left SAE 200 Mode 4 Down right SAE 1032 Lowest SAE Best prediction Being currently predicted SAE Sum of Absolute Errors Figure 3 4 Intra mode decision based on lowest SAE 10 Inter Mode Decision MB modes for inter prediction in AVS China P2 Each 16x16 MB can have 1 to 4 motion vectors representing it Figure 3 5 Inter mode decision 3 The MB type that provides the best prediction is chosen after SAE 4 Machine Learning Data mining Machine learning Supervised learning Classification Decision tree classifiers C4 5 J48 algorithm WEKA Waikato Environment for Knowledge Analysis data mining tool open source software popular suite of machine learning software written in Java developed at the University of Waikato Mode decision complexity reduction Extract Encoder mode decisions offline Extract Attributes offline Extract Attributes online C4 5 J48 classifier Generate decision trees Convert to if else statements Replace complex mode decision by if else statements to determine the mode Fig 4 1 Low complexity Encoder using machine learning Classify each MB with the correct mode Attribute Mean of each current 16x16 MB Variance of each current 16x16 MB Mean of 16 4x4 blocks within the current 16x16 MB Variance of this 16 means for the current 16x16 MB Variance of 16 4x4 blocks within the current 16x16 MB Mean of each 16x16 residue MB Variance of each 16x16 residue MB Mean of 16 4x4 blocks within the 16x16 residue MB Variance of this 16 means for the 16x16 residue MB Variance of 16 4x4 blocks within the 16x16 residue MB Edge detection for 16x16 MB yes no Edge detection for 16x16 residue MB yes no Edge direction for current 16x16 MB Edge direction for 16 4x4 blocks within the current 16x16 MB Edge direction for 16x16 residue MB Edge direction for 16 4x4 blocks within the 16x16 residue MB Total 100 attributes for each macroblock Edge Detection Roberts Prewitt Quantize edge intensity to 0 1 set a threshold white pixel denotes strong edge Sobel Canny Canny s edge detection algorithm is computationally more expensive compared to Sobel Prewitt and Robert s operator However the Canny s edge detection algorithm performs better than all these Edge Detection Mode Decision Mode 4 Down right Mode 4 Down right SAE Mode 0 Vertical mode Mode 3 Down left Mode 1 Horizontal Feature Extraction an Example Fig 4 2 Extracting features of MB adjacent to current block in intra prediction 5 Summary Within AVS China Part II video codec an area for improvement is identified which is the mode decision computation which is a very computationally expensive process An algorithm is proposed to reduce this computation by replacing the complex mode decision by a set of if else statements that can fairly accurately mimic the existing mode decisions This set of if else statements is generated from a decision tree which is one of the classification methods in data mining In this proposal only one algorithm is outlined targeting intra mode decision This approach along with some more modifications can be applied to inter mode decisions also It is proposed to implement this and evaluate its impact on reducing the implementation complexity of AVS China part II video codec 6 References 1 AVS Video Expert Group Information technology Advanced coding of audio and video Part 2 Video AVS1 P2 JQP FCD 1 0 Audio Video Coding Standard Group of China AVS Doc AVS N1538 Sep 2008 2 AVS Video Expert Group Information technology Advanced coding of audio and video Part 3 Audio Audio Video Coding Standard Group of China AVS Doc AVS
View Full Document
Unlocking...