フレーム内再構成型超解像の領域適応処理による高画質化の検討

摘要

フレーム内処理の再構成型超解像方式を，エッジ部に加えてテクスチャ部や平坦部でも高画質化させる．従来のフレーム内再構成超解像方式は，輝度変化が被写体の輪郭に沿って連続する性質である自己合同性に着目し，エッジ部分において高い鮮鋭感が得られる高解像度化を行う．今回，まず，入力画像をエッジ部／テクスチャ部／平坦部に分割する．続いて，エッジ部では入力画像と自己合同性から生成した条件，テクスチャ部では入力画像から生成した条件を満たすように，フィルタ方式で拡大した画像を修正することで高解像度化する．従来のエッジ部分での鮮鋭感向上に加えて，テクスチャ部分での明瞭感の向上や，平坦部でのノイズの増幅を抑えることができ，17名の被験者による3次畳み込み内挿法との一対比較で約60［％］の採択率を得た．%We propose an improved one frame reconstruction-based super-resolution method which increases image quality even in texture and flat regions in addition to edge regions. The conventional method utilizes the image self-congruence that the intensities along edges are continuous, and improves the edge sharpness. In the proposed method, an input image is segmented and classified to a texture region, a flat region and an edge region. To the images of the texture region, the input image magnified by the filter method is modified so as to satisfy the constraint generated from the input image. To the images of the edge region, the conventional one frame reconstruction-based super-resolution method is applied. To the images of the flat region, the input image is magnified by the filter method. The subjective evaluation by 17 people proved that the reconstructed image by the proposed method has better clearness in the texture regions and the less noise in the flat regions in addition to sharp edges than the cubic convolution method.