LMS-like two dimensional adaptive filter with t-distribution assumption and nonsymmetrical half-plane support

This paper will focus on proposing a new objective function of the two-dimensional adaptive filters. The objective function is derived by assuming that the obtained error signal is identically and independently distributed with t-distribution probablity density function. Small portions of error signals with large amplitude is given a small weighting factor and large portions of error signals with small amplitude is assiggned a large weighting factor. By doing so, the effect of large amplitude error signal is suppressed and the adaptation step is less affected. The simulation results for image enhancement show that when the input is impulsive noise contaminated images, the obtained processed image by using t-distribution with small α degrees of freedom is much better than that when large α is applied.