Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 22Slide 23Slide 24Slide 25Modified advanced image codingZhengbing Zhang Electronics and Information College, Yangtze UniversitySupervisor: Dr K.R. RaoElectrical Engineering Department, University of Texas at ArlingtonOutline1. Introduction2. JPEG-Baseline3. JPEG 20004. Advanced Image Coding5. Modified Advance Image Coding(M-AIC)6. Simulations7. Conclusions and Future Work1. Introduction•JPEG[1] has played an important role in image storage and transmission since its development. •JPEG provides very good quality of reconstructed images at low or medium compression but it suffers from blocking artifacts at high compression. •Several papers [2]~[7] have been published to improve the performance of DCT-based image compression. •In his website[8], Bilsen provides an experimental still image compression system known as Advanced Image Coding (AIC) that performs much better than JPEG and close to JPEG-2000[10].2. JPEG-Baseline(a) Encoder(b) Decoder3. JPEG 2000•Based on wavelet transform•Context Coding Algorithm: EBCOT (Embedded Block Coding with Optimal Truncation)•Context-based Arithmetic Entropy Coding•This simulation disables tiling and scalable mode•Reference software[10]: JasPer v 1.900.14. Advanced Image Coding (a) Encoder [8] (b) Decoder [8]Advanced Image Coding It is a still image compression system which is a combination of H.264 and JPEG standards.Features:No sub-sampling- higher quality / compression ratios9 prediction modes as in H.264Predicted blocks are predicted from previously decoded blocksUses DCT to transform 8x8 residual block instead of transform coefficients as in JPEGEmploys uniform quantizationUses floating point algorithmCoefficients encoded in scan-line orderMakes use of CABAC similar to H.264 with several contexts5. M-AIC(a) M-AIC Encoder(b) M-AIC DecoderBGRCrCbYCCMode Selectand StoreBlockPredictmodeYY, Cb, Cr Blks++Pred BlkFDCTQZZHuffAACQ1IDCT+TableResResDec YDecYDecCbDecCrPredictor ModeEncBGRCrCbYICCBlockPredictY,Cb,Cr Blks++Pred BlkIDCTQ1IZZIHuffAADTableResModeDecand StoremodeDecYDecCbDecCrCC - color conversion, ICC - Inverse CC, ZZ – zig-zag scan, IZZ – inverse ZZ, AAC – adaptive arithmetic coder, AAD – AA decoder.Color ConversionY = 0.299R + 0.587G+ 0.114BCb=-0.169 R - 0.331G +0.5 BCr= 0.5 R - 0.419G - 0.081 B R=Y+ 1.402CrG=Y - 0.344Cb-0.714CrB=Y+ 1.772Cb YCbCr format is 4:4:4. The color conversion method same as in JPEG reference software [9] is used.Prediction Modes[8]Mode 0: Vertical Mode 1: Horizontal Mode 2: DCMode 3: Diagonal Down-LeftMode 4: Diagonal Down-RightMode 5: Vertical-RightMode 6: Horizontal-Down Mode 7: Vertical-Left Mode 8: Horizontal-UpPrediction Modes (contd.)•Determine only when coding each Y block •By full search among the 9 modes•minimize the prediction error with Sum of Absolute Difference•The selected prediction mode is stored & used for blocks in Y, Cb and Cr. •ModeEnc encodes selected prediction modes with a variable length algorithm.Encode the prediction residual•The prediction residual (Res) is transformed into DCT coefficients with floating point DCT. •DCT coefficients are uniformly scalar-quantized: same QP for all the DCT coefficients of Y, Cb and Cr. •zig-zag scan•Encode 64 coefficients of a block with the same algorithm for the AC coefficients in JPEG[1][9]. •Use the Huffman table for AC coefficients of chrominances recommended in baseline JPEG [1][9].File Format•stream header : 11 bytes (format flag, version, QP, image width, image height, pixel depth, code size of the compressed modes). •stream order: header, code of prediction modes, Huffman codes of Y-Res, Cb-Res and Cr-Res. • An adaptive arithmetic coder [12][13]: input byte-by-byte from the compressed stream; output finally compressed result.M-AIC CodecM-AIC Codec6. Simulations•Performance comparisons with bit-rate vs PSNR•Original and compressed Lena image with different methodsTest images(a) Lena 51251224 (b) Airplane 51251224 (c) Couple 25625624 (d) Peppers 512 51224 (e) Splash 51251224 (f) Sailboat 51251224Performance comparisons with bit-rate vs PSNR (a) Lena (512x512x24) (b) Airplane (512x512x24)(c) Couple (256x256x24) (d) Peppers (512x512x24)0 0.2 0.4 0.6 0.8 1 1.2 1.41820222426283032343638Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG20000 0.5 1 1.5 2 2.5152025303540Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG20000 0.2 0.4 0.6 0.8 1 1.2 1.418202224262830323436Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG20000 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8101520253035Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG2000Performance comparisons with bit-rate vs PSNR(contd.)(e) Splash (512x512x24)(f) Sailboat (512x512x24)0 0.5 1 1.5 2 2.5 3 3.5 41015202530354045Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG20000 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.814161820222426283032Bits Per PixelPSNR dBAICM-AICJPEG-RefJPEG2000Original and compressed Lena image with different methods (a) Original Lena (51251224) (b) AIC: 0.22bpp, PSNR=28.84dB (c) JPEG2000: 0.22bpp, PSNR=29.57dBCompressed Lena image with different methods(contd.)(d) M-AIC: 0.22bpp, PSNR=29.02dB (e) JPEG: 0.22bpp, PSNR=24.29dBCompressed Lena image with different methods(contd.)(f) AIC: 0.15bpp, PSNR=27.29dB (g) M-AIC: 0.15bpp, PSNR=27.43dB (h) JPEG: 0.16bpp, PSNR=14.05dBConclusions and Future Work•M-AIC performs much better than baseline JPEG, close to AIC and JPEG-2000, and a little bit better than AIC at some low bit rate range. •Replace the Huffman coder and AAC with CABAC•Replace floating point DCT with integer DCT•Try more prediction modesReferences1. W. B. Pennebaker and J. L. Mitchell, JPEG still image data compression standard, Van Nostrand Reinhold, New York, 1993.2. A. Gupta et al., “Modified runlength coding for improved JPEG performance,” Intl. Conf. on Information and Communication Technology,2007, pp. 235 – 237, Dhaka, Bangladesh, March 2007.3. G. Lakhani, “DCT coefficient prediction for JPEG image coding,” IEEE Int. Conf. Image Processing, 2007, vol. 4, pp. IV-189 – IV-192, Oct. 2007. 4. C. Wang, et al., “An improved JPEG compression algorithm based on sloped-facet model of image segmentation,” Intl. Conf. on Wireless Communications,
View Full Document
Unlocking...