Aproximações da DCT de comprimento 16 com baixa complexidade aritmética para compressão de imagens

Abstract

Image compression is a technique that aims an image storage space reduction in secondary memory. One cenary where this technique can be used is the image transmission in a network. In this process, a compressed image can be transmitted through this network – or data bus – in less time if compared with the same image with no compression. Complementing this cenary, one can imagine that image capture devices have low processing and storage power. Thus, it is important that these devices can compress and, after, transmit images through a network or data bus with the lowest possible computational cost and, consequently, in a shorter time. Usualy, discrete transforms – especially those with trigonometric core – are used in image compression. The well known discrete cosine transform (DCT) is used in important compression standards as JPEG and MPEG-1. However, when the processing capacit is restrict, low computational cost approximations of this transform are presented as interesting options. There exist many DCT approximations in literature – especially 8-point transforms because this block lenght is applied in usual standards. Indeed, other compression standards, as the recent HEVC, use not only the 8-point DCT but also the 4, 16 and 32-point DCT. In this context, this work does a review of all 16-point DCT approximations found in literature so far, as well presents a new transform of same block lenght. The proposed transform’s fast algorithm is multiplication free and have the lowest computational cost archived in the literature. To evaluate the proposed transform, in contrast to those found in literature, image quality, DCT similarity and coding gain metrics are implemented. Consequently, it is shown that the proposed transform presents good results, low computational cost and has potential to be efficiently implemented in software and hardware.