Contrast-Enhanced Black and White Images Hua Li and David Mould UNC Wilmington and Carleton University Presented by Ling Xu
The Problem of Tone Reduction It is difficult to obtain a clear segmentation without user user Bernard Levine
Related Work Thresholding methods Details are missing. Region boundaries have isolated pixels. Our method preserves the details very well.
Related Work in Non-Photorealistic Rendering (NPR) --- Region-Based Methods [XKM07, MG08, XK08, RL10, MZZ10] By Xu and Kaplan [XK08] Our method Highly rely on the results from segmentation methods
Related Work in Non-Photorealistic Rendering (NPR) --- Filter-Based Methods [Win11,WKO12] Our method By Winnemöller et. al [Win11,WKO12] Sharp corners are generally missing, and unwanted edge extensions may depict the content wrongly.
Our Algorithm Inspired by Stick filtering Several stages: 1.initial global blurring 2.a conversion from a color image to an enhanced and smoothed greyscale image 3.binary thresholding 4.final cleanup -- further smoothing and removal of small clusters
Step 2: a conversion from a color image to an enhanced and smoothed greyscale image For each pixel, running statistical analysis on a few approximated directions Contrast is increased along the direction of maximum response, amplifying local tonal differences. Or smoothing is applied when the pixel is in an area with little texture or few edges. This conversion is iterated for a couple of times.
Our Algorithm Original image Contrast-enhanced without smoothing Contrast-enhanced and smoothed Final effect after thresholding and removal of small regions
Core Ideas about Contrast amplifying local tonal differences iteratively Original
Statistical Analysis on each pixel For each pixel, we approximate n directions and study the statistical values to decide if a pixel needs to be darkened or lightened. The direction of maximum difference between the average intensity of a stick and the average intensity of the local area. Chosen
Noise Suppression Using Global and Local Smoothness Original image Contrast-enhanced without smoothness Contrast-enhanced and smoothed
Variations of Our Algorithm
Variations of Our Algorithm
Variations of Our Algorithm
Our Method Xu and Kaplan’s Method
Comparisons
Comparisons
20 Our method By Winnemöller et. al [Win11,WKO12]
Advantages o Automatic o Simple implementation o High quality in structural preservation o Control over the detail and the abstraction
Limitations o Demands multiple iterations slower than the XDOG method o Demands the configuration of a set of parameters the same as previous methods o Demands further investigation on the variations
More Results
Acknowledgements Thanks to Ling Xu for the presentation. Thanks to Holger Winnemöller for sharing the images. Thanks to the reviewers for the comments. Thanks to Peter Selinger for the Potrace tool.
Questions? Please send your questions to Hua Li directly.