LEARNING IMAGE FEATURES TO ENCODE VISUAL INFORMATION LIFEVISION

LEARNING IMAGE FEATURES TO ENCODE VISUAL INFORMATION (LIFE-VISION) Kick-off meeting 28/01/2013

VISUALLY LOSSLESS CODING Visibility thresholds LIFE-VISION

VISUALLY LOSSLESS CODING Visibility thresholds LIFE-VISION

VISUALLY LOSSLESS CODING Visibility thresholds LIFE-VISION

VISUALLY LOSSLESS CODING Visibility thresholds LIFE-VISION

VISUALLY LOSSLESS CODING Visibility thresholds LIFE-VISION

VISUALLY LOSSLESS CODING Current methods: focused on the encoding process JPEG 2000 codestream Research purpose: extract a visually lossless image from a JPEG 2000 codestream LIFE-VISION

EMBEDDED QUANTIZATION 0 W LIFE-VISION

EMBEDDED QUANTIZATION W 0 uniform scalar deadzone quantization (USDQ) LIFE-VISION

EMBEDDED QUANTIZATION 0 emit 1 W bitplane coding (BPC) LIFE-VISION

EMBEDDED QUANTIZATION 0 0 0 1 1 W bitplane coding (BPC) LIFE-VISION

EMBEDDED QUANTIZATION 0 0 1 0 1 W bitplane coding (BPC) WHY IS USDQ+BPC SO POPULAR ? 1) is convenient for current hardware 2) achieves competitive coding performance 3) many works aimed to improve efficiency IS USDQ+BPC OPTIMAL FOR LOSSY IMAGE CODING ? LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) 0 emit 1 W T 1 LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) 0 0 T 1 1 W T 2 LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) 0 0 1 T 3 W T 1 T 2 LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) 0 W LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) W 0 different thresholds: quantizers with K stages: LIFE-VISION

EMBEDDED QUANTIZATION W 0 GENERAL EMBEDDED QUANTIZATION (GEQ) W 0 different thresholds: quantizers with K stages: Example: LIFE-VISION