Presentation is loading. Please wait.

Presentation is loading. Please wait.

Tuning JPEG2000 Image Compression for Graphics Regions

Similar presentations


Presentation on theme: "Tuning JPEG2000 Image Compression for Graphics Regions"— Presentation transcript:

1 Tuning JPEG2000 Image Compression for Graphics Regions
Serene Banerjee and Brian L. Evans Department of Electrical Engineering The University of Texas at Austin 04/08/02 Tuning JPEG2000 for Graphics

2 Tuning JPEG2000 for Graphics
Objective JPEG2000 Provides higher compression rates than JPEG for same visual quality for natural images Adopted in cellular phones, PDAs, Web, and printers Problem Formatted text and graphics data may be present At same bit-rate JPEG2000 has lower compression rates than lossless GIF / PNG coders for graphics data Goal: Tune JPEG2000 encoder for graphics 04/08/02 Tuning JPEG2000 for Graphics

3 Natural Image Results 512 x 512 x 24 bits per pixel ‘Lena’ at 80:1 compression 512 x 512 x 24 bits per pixel ‘Peppers’ at 80:1 compression 04/08/02 Tuning JPEG2000 for Graphics

4 Contrast sensitivity function (possible frequency weighting)
Visual Artifacts 4-levels of wavelet decomposition JPEG2000 Wavelet transforms Frequency weighting Truncates high frequency coefficients Zero noticeable distortion due to truncation is not valid for graphics data Contrast sensitivity function (possible frequency weighting) 04/08/02 Tuning JPEG2000 for Graphics

5 Visual Frequency Weighting
Lowpass CSF weighted wavelet coefficients Fixed or varying weights based on image content Function of viewing distance and image size Optimized tables for natural images 04/08/02 Tuning JPEG2000 for Graphics

6 Visual Frequency Masking
Self-contrast masking Wavelet coefficients raised to alpha power Neighborhood masking Distortion measure for each pixel is function of neighbors’ distortion Pointwise extension Above methods combined 04/08/02 Tuning JPEG2000 for Graphics

7 Tuning JPEG2000 for Graphics
Post-Processing Algorithm From original image pixel-to-pixel difference computed for all edges Minimum and maximum threshold value obtained In decompressed image disregard edges outside threshold values Lowpass filter image 04/08/02 Tuning JPEG2000 for Graphics

8 Graphics Image Results
JPEG2000 compressed (at 80:1) graphics images have ringing distortion, whereas PNG / GIF encodes losslessly at same bitrates 04/08/02 Tuning JPEG2000 for Graphics

9 Frequency weighted image
Results Ringing distortion is correlated with image edges Distortion quantified with inverse of correlation between error and edges Measure unable to represent visual judgment Original image Frequency weighted image Visual masked image Post-processed image 04/08/02 Tuning JPEG2000 for Graphics

10 Tuning JPEG2000 for Graphics
Conclusions Abrupt changes in pixel values of graphics make JPEG2000 compression less efficient Post-processing method Produces better visual results for graphics Is not standard compliant Adds to decoder complexity (3 x 3 filter) Visual weighting and masking methods Produces some improvement for graphics Standard compliant Requires image content segmentation at encoder 04/08/02 Tuning JPEG2000 for Graphics


Download ppt "Tuning JPEG2000 Image Compression for Graphics Regions"

Similar presentations


Ads by Google