Coiflets for range image segmentation

Abstract

In this paper, we propose a method for accelerating range image segmentation process which uses techniques of region-growing based on functions approximation and local neighborhood property. This method uses multiresolution wavelet transforms, where the used wavelets are Coiflets. The interesting property of this kind of wavelets is the interpolating characteristic of their associated scaling functions. This characteristic is due to the fact that both Coiflets and their scaling functions moments vanish. An overview on the proposed segmentation scheme gives the following description. First, input data is compressed to a fixed lower resolution (LR), then the partitioning process is applied on the compressed image. The result of this process is “projected” to the initial resolution (IR) and a global error criterion (GEC) is evaluated. The required result is obtained by increasing LR -if necessary- until the GEC is respected. Encouraging results are obtained for several range images.