Learned Focused Plenoptic Image Compression with Microimage Preprocessing and Global Attention

Focused plenoptic cameras can record spatial and angular information of the light field (LF) simultaneously with higher resolution relative to traditional cameras, which facilitate various applications in computer vision. However, existing image compression methods present ineffectiveness captured images due complex micro-textures generated by microlens relay imaging long-distance correlations among microimages. In this paper, a lossy end-to-end learning architecture is proposed compress focused efficiently. First, data preprocessing scheme designed according principle remove sub-aperture ineffective pixels recorded align microimages rectangular grid. Then, global attention module large receptive capture correlation feature maps using pixel-wise vector computed resampling process. Also, new dataset consisting 1910 content depth diversity built benefit training testing. Extensive experimental evaluations demonstrate effectiveness approach. It outperforms intra coding HEVC VVC an average 62.57% 51.67% bitrate reduction on 20 preprocessed images, respectively. it achieves 18.73% saving generates perceptually pleasant reconstructions compared state-of-the-art methods, benefits greatly. The code are publicly available at https://github.com/VincentChandelier/GACN .