Progressive contour coding in the wavelet domain

This paper presents a new wavelet-based image contour coding technique, suitable for representing either shapes or generic contour maps. Starting from a contour map (e.g. a segmentation map or the result of an edge detector process), a unique one-dimensional signal is generated from the set of contour points. Coordinate jumps between contour extremities when under a tolerance threshold represent signal discontinuities but they can still be compactly coded in the wavelet domain. Exceeding threshold discontinuities are coded as side information. This side information and the amount of remaining discontinuity are minimized by an optimized contour segment sequencing. The obtained 1D signal is decomposed and coded in the wavelet domain by using a 1D extension of the SPIHT algorithm. The described technique can efficiently code any kind of 2D contour map, from one to many unconnected contour segments. It guarantees a fully embedded progressive coding, state-of-art coding performance, good approximation capabilities for both open and closed contours, and graceful visual degradation at low bit-rates.