基于ASAM注意力机制的光场图像压缩方法

为了研究光场图像的空间信息和相似角度信息之间的差异性，提高光场图像的传输效率，提出了一种基于端到端网络的角度空间注意力模型（ASAM）注意力机制的光场图像压缩方法 .以卷积块注意力模型（CBAM）的注意力机制为基础，增强了相对角度特征，提高了压缩编码效率.稀疏图像采用H.266/VVC视频编解码器进行压缩，通过子孔径图像（SAI）网络恢复编码后的图像.结果表明，与现有的光场图像压缩方法相比，所提出的光场图像压缩方法具有较高的图像压缩性能，Bj?ntegaard-Delta比特率（BD-BR）降低了52.30%,Bj?ntegaard-Delta峰值信噪比（BD-PSNR）提高了3.33 dB.

Light field image compression method based on ASAM attention mechanism

A light field image compression method based on an end-to-end network with angular-spatial attention model (ASAM) attention mechanism was proposed in order to study the difference between the spatial information and similarity angular information of light field images, which improved the transmission efficiency for light field images. The method was constructed based on the attention mechanism of convolutional block attention module(CBAM), and the relative angle feature was enhanced to improve the compression coding efficiency. The sparse images were compressed by H.266/VVC video coding. The coded images were recovered by a subaperture image (SAI) recovery network. The result showed that the proposed light field image compression method had higher image compression performance with BD-BR value reduced by 52.30%, and BD-PSNR improved by 3.33 dB compared with the existing optical field image compression methods.