Subdivision wavelet with stochastic coefficients

Wavelet-based image processing techniques such as de-noising and multiresolution representation typically model the wavelet coefficient as independently or jointly fixed numbers. These models are unrealistic for some real world signals. In the paper, we develop a new framework for subdivision wavelet based on 4-point interpolatory subdivision scheme of N. Dyn and D. Levin. The new subdivision wavelet is found by closed pentagon and its coefficients are stochastic with some range. To demonstrate the utility of the new framework, the comparisons of reconstruction images from initial images via subdivision wavelet and subdivision are executed. Experiments show that the error of signal-to-noise between them is less than 0.1 when the parameters are restrained in the range of plusmn0.1. Further results show that the new framework is superior to other two methods.