Computation-aware algorithm selection approach for interlaced-to-progressive conversion

摘要

We discuss deinterlacing results in a computationally constrainednand varied environment. The proposed computation-aware algorithmnselection approach u0001CASAu0002 for fast interlaced to progressivenconversion algorithm consists of three methods: the line-averaging u0001LAu0002nmethod for plain regions, the modified edge-based line-averagingnu0001MELAu0002 method for medium regions, and the proposed covariancebasednadaptive deinterlacing u0001CADu0002 method for complex regions. Thenproposed CASA uses two criteria, mean-squared error u0001MSEu0002 and CPUntime, for assigning the method. We proposed a CAD method. The principlenidea of CAD is based on the correspondence between the high andnlow-resolution covariances. We estimated the local covariance coefficientsnfrom an interlaced image using Wiener filtering theory and thennused these optimal minimum MSE interpolation coefficients to obtain andeinterlaced image. The CAD method, though more robust than mostnknown methods, was not found to be very fast compared to the others.nTo alleviate this issue, we proposed an adaptive selection approach usingna fast deinterlacing algorithm rather than using only one CAD algorithm.nThe proposed hybrid approach of switching between the conventionalnschemes u0001LA and MELAu0002 and our CAD was proposed to reducenthe overall computational load. A reliable condition to be used for switchingnthe schemes was presented after a wide set of initial training processes.nThe results of computer simulations showed that the proposednmethods outperformed a number of methods presented in thenliterature.