Deep Image Compression using BINet Andr Nortje 18247717@sun.ac.za - PowerPoint PPT Presentation

Deep Image Compression using BINet André Nortje 18247717@sun.ac.za Prof. Herman Engelbrecht hebrecht@sun.ac.za Dr Herman Kamper kamperh@sun.ac.za � 1

The Standard Approach • Image Compression • JPEG : DCT -> Quantisation Tables -> Entropy Coding [3]. • WebP: like JPEG but with Linear in-painting [1]. • Problem • Standard codecs su ff er from arduous hand-tuning of quantisation parameters and lack of end-to-end system optimisation. • Solution • Deep Learning ;)

Deep Image Compression Basic Architecture B i Encoder Decoder n

Why Patch Based Schemes are Avoided Block Artefacts

BINet : BINary INpainting Network

Masked BINet: Is inpainting from binary codes possible?

Inpainting Complexity

Does the ability to inpaint aid compression?

PSNR Compression Curve

Current and Future Work • We’ve created a Proof Of Concept … now: • Separate inpainting and patch decoder seems to lead to greater gains. • Training the Google Recurrent Model [7] to become a BINet, attempt beating current S.O.T.A • Application of BINet to Error Correction/ Detection systems.

References 1. WebP : A new Image Format for the Web (2016) [Website]. Available at https://developers.google.com/speed/webp/docs/compression (Accessed: 2018-07-30) 2. Cisco (2015), Cisco Visual Networking Index: Forecast and Methodology, 2016-2021 [Journal Article]. Available at https://www.cisco.com/c/en/us/ solutions/collateral/service-provider/visual-networking-index-vni/complete-white-paper-c11-481360.html#_Toc484813971 (Accessed: 2018-08-23) 3. G. Wallace (1991), The JPEG Compression Standard, Communications of the ACM, Available at http://www.ijg.org/files/Wallace.JPEG.pdf (Accessed: 2018-07-31) 4. Iain E. Richardson, The H.264 Standard Video Compression Standard 2nd Edition, Vcodex Limited Uk, Available at https://files.cnblogs.com/files/ irish/The_H.264_advanced_video_compression_standard.pdf (Accessed: 2018-07-31) 5. T. Raiko, M. Berglund, G. Alain et al. (2014), Techniques for Learning Binary Stochastic Feedforward Neural Networks, Available at https://arxiv.org/ pdf/1406.2989.pdf (Accessed: 2018-07-31) 6. G. Toderici, S. O’Malley, S. Hwang et al. (2015), Variable Rate Image Compression with Recurrent Neural Networks, Available at http://arxiv.org/abs/ 1511.06085 (Accessed: 2018-07-31) 7. G. Toderici, D. Vincent, N. Johnston (2016), Full Resolution Image Compression with Recurrent Neural Networks, Available at https://arxiv.org/abs/ 1608.05148 (Accessed: 2018-01-25)