Casamento Aproximado de Padrões Multiescala com Segmentação Flexível e Treino do Dicionário

Abstract

In this paper we present a new segmentation method for the MMP (Multidimensional Multiscale Parser) algorithm, allied with a dictionary training procedure. These improvements allowed MMP to outperform state-of-the-art transform-based encoders like JPEG2000 and H.264/AVC, for both smooth and non-smooth images, at low to medium compression ratios. The new approach allowed MMP to exploit the images’ structure in a more effective way, increasing significantly its performance for smooth images. Additionally, the new techniques do not compromise the results achieved for non-smooth images, where MMP already outperformed transform-based algorithms. Keywords— Image coding, pattern matching, image segmentation, image processing