HCLE: Codificador de Light Fields para altas taxas de compressão com predição baseada em Optical Flow e Phase Correlation

Abstract

Light Field is a very promising technology to represent digital visual signals. With Light Fields it is possible to capture and represent in a more faithful fashion the captured scenes compared to previous imaging technologies. The Light Fields have a extense range of applications, variyng from medical to immersive and virtual reality applications. However, it was only a few years ago that technologies emerged to make possible to capture Light Fields, like Lytro Illum camera. Despite the many challenges related to Light Fields manipulation, mainly related to compression, there are many works focusing on Light Fields compression to make it a feasible technology. Some works use Light Fields as videos to compress it in video encoders, to use the already implemented tools present in these encoders. Others develop new techniques in order to explore unique inherent characteristics of Light Fields. However, there are few works that utilize prediction to compress Light Fields. Thus, this work proposes a Light Field Encoder for High-Compression Rate susing Optical Flow and Phase Correlation-based Prediction (HCLE). The proposed encoder reaches a compression rate of 99.9996% in comparison to the original Light Field, besides removing visual artifacts that lenslet Light Fields originally have. Comparing to the related work, the HCLE reaches a compressiont rate of 35.22% for each SAI, generating all 225 lenslet Light Fields SAIs. To assess the Light Field visual quality, it was used the BRISQUE method, reaching a visual enhancement of 56.21%. It was also proposed a scheme to compensate prediction error utilizing the HCLE along with the HEVC, making the proposed solution more competitive to related works.