Real-time spatially adaptive image restoration using truncated constrained least squares filter

Abstract

A finite impulse response (FIR) filter design method is presented by truncating the constrained least squares filter for real-time, spatially adaptive image restoration. The proposed method truncates the original constrained least squares image restoration filter using the Maxwell-Boltzmann distribution kernel. For the edge preserving image restoration, the orientation of local edge is analyzed based on the covariance matrix, and the edge orientation-adaptive restoration filters are generated. The reduced size of the FIR type restoration filter makes hardware implementation easier for real-time image enhancement. Experimental results show that the proposed method provide more detail and less restoration artifacts than existing methods. As a result, the proposed restoration filter can be applied to realtime image enhancement systems, such as high-definition televisions and video surveillance systems.