Spatial Feature Based Video Scaling Scheme and its FPGA Implementation for Video Standards Conversion

Abstract

certain conditions. We present an FPGA implementation of the In many digital image/video processing applications resolution proposed algorithm. The resulting hardware design is tested for 1 a standards conversion application where 480 x 720 progressive enhancement naturally arises as a problem ofhigh practical value. amst are caledsto720ixe1280 progressive Typically, increasing spatial resolution through modifications in ond. the imaging system is not a feasible option, and post-processing Rans algorithms designed to enhance resolution of the acquired im-Rvanced scaling technques that can recover the missing or aliased age/video signal prove beneficial. In this work, we analyze re-high fre quenets. Singlerameresoluton enhanecent work on pixel classification based resolution enhancement, ment teques canest Singl highmfrequeny ncnamely, resolution synthesis, and discuss its applicability to low nents tocanimied tnt throughespatiallyhaaptiveqfilteringpa complexity customer grade display systems. In the light of our oeoresolution use of prior information. The main improvement offered by sinobservatsios, proin moutfcer schort-oimigof gle frame resolution enhancement is observed around edges and syrmntes anderperope mondifieds.W cementoimFPro itsmpr-textured areas. Compared to the results obtained by LSI scaling formance under certain conditions. We present an FPGA imfles uha iui neplto obndwt nhre plementation of the proposed algorithm, and provide a compu-filteri, such as 'teresotion synthesi arith tational complexity analysis. The resulting hardware design is f . s . t can offer much smoother, continuous edges with sharp) transitested for a standards conversion application where 480 x720 ca.fe uhsote,cotnosegswt hr tastestd fr astanard coverion pplcaton here480x 70 tons and remove the blurry look from textured areas. There are progressive frames are scaled to 720 x 1280 progressive at 60 ' at least two well-known single-frame resolution enhancement algorithms that utilize prior information in this format, namely, Index Terms-resolution enhancement, FPGA implemen-resolution synthesis proposed by Atkins et. al. [1] and exampletation, scaling, resolution synthesis, standards conversion. based super-resolution by Freeman et. al. [2]. Resolution synthesis (RS) is based on pixel classification and adaptive linear filtering, and is computationally less demanding compared to example-based super-resolution. Since, our goal is to design a For many digital image/video processing applications increas-low-complexity resolution enhancement method that can be iming the spatial resolution is not only desirable but also highly plemented in the next generation display systems, we focus on beneficial. At higher resolution, TV pictures look more natural the RS algorithm. and pleasing to the eye, security cameras can offer better iden-. The rest of the paper is organized as follows. Section 2 protification, and satellite imagery can be interpreted with higher vides a brief overview of the RS algorithm, discusses its appliaccuracy. As s...