Rate-Distortion Optimal Skeleton-Based Shape Coding

Haohong Wang, Aggelos K. Katsaggelos, and Thrasyvoulos N. Pappas

Image Processing, 2001. Proceedings. 2001 International Conference on , Oct 2001

Outline

Introduction Skeleton-Based Shape Representation Lossy Coding of Shape Data Experimental Results Conclusions

Introduction

Shape coding– Block-based– Contour-based

Chain Code

Baseline

Vertex-based (polygon/B-spline approximation)

The binary shape of an object is defined by

Skeleton-Based Shape Representation The basic idea :

– To represent an object by one or more 2D curves (skeletons with associated distance data)

Skeletons– The center of the object in the horizontal direction

Distance data– The associated distance from the boundary

Skeleton-Based Shape Representation

Skeleton-Based Shape Representation More flexibility

– Different transform and compression methods for each data set, according to their characteristics.

The skeletons can be used for the estimation of the object motion in the Inter-mode

Coding of Shape Data

Different characteristics– The object has near axial symmetry => the

skeleton will be smooth– The two boundaries are nearly parallel =>

the distance function will be smooth The experimental results verify that the

decoupling provides a compression advantage.

Coding of Shape Data

Polygonal approximation

Lossy Coding of Shape Data

To solve the problem :

max1010 ),...(:),,...(min RPPRtosubjectPPD NN

max1010 ),...(:),,...(min DPPDtosubjectPPR NN

Lossy Coding of Shape Data

Distortion function– The sum of the absolute value of the errors

Lossy Coding of Shape Data

Lossless skeleton and lossy distance data coding

sdN RRRtosubjectppD max10 :),,...(min

Lossy Coding of Shape Data

Lossy coding of both skeleton and distance data

To minimize

max1010 :)],,...(),...([min RRRtosubjectqqDppD dsMdNs

)( dsds RRDD

pixelsInteriorofNumber

errorinpixelsofNumberDMPEG 4

Lossy Coding of Shape Data

Experimental Results

polygonspline

Conclusions

By decoupling the shape object data into the skeleton and distance data, the scheme reduces their correlation.

The approach together with polygonal approximation of the skeleton and the distance data results in a significant improvement in rate-distortion efficiency.

Conclusions

Which is the suitable algorithm for compressing the skeleton and the distance data?

The skeleton and distance data can be predicted from the previous frame.