Multi-scale boundary representation using wavelet detection and arithmetic chain coding

Abstract

This project involves the development of a multi-scale boundary representationusing wavelet detection and an improved chain-coding algorithm to codeboundary information.Classical gradient-based edge detectors (Roberts, Sobel, etc.) which use localpixel intensity changes to locate edges, are not able to distinguish edges ofdifferent significance. The proposed wavelet-based edge detection algorithmanalyses the interaction of the 1-D wavelet transform coefficients (acrosstransform scales) along image rows and columns to identify multi-scale edges ofdifferent significance. Next, the algorithm re-classifies the edge informationaccording to the 2-D correlation of the edge pixels. Edge pixels having strongerspatial correlation with neighbouring edge pixels would be promoted to a higherlevel of significance while those having weaker correlation with neighbouringedge pixels would be demoted. With this, the multi-scale edge detector isimplemented.The proposed Arithmetic-based Chain Coder encodes the boundary informationby applying arithmetic coding into chain-coding. The experimental results showpositive improvement in coding performance (by 29%) when compared toconventional chain coding or scanline arithmetic coding.