基于尺度特征融合的极低比特率三维复杂图像无损压缩  

Lossless Compression of 3D Complex Images with Very Low Bit Rate Based on Scale Feature Fusion

作　　者：黄梅娟[1] 王永梅[2] HUANG Meijuan;WANG Yongmei(School of Information Engineering,Anhui Business and Technology College,Hefei 231131,China;School of Information and Artificial Intelligence,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]安徽工商职业学院信息工程学院,合肥231131 [2]安徽农业大学信息与人工智能学院,合肥230036

出　　处：《海南热带海洋学院学报》2024年第5期59-66,共8页Journal of Hainan Tropical Ocean University

基　　金：安徽省职业与成人教育学会教育教学研究规划课题(AZCJ2023113);安徽省高校优秀拔尖人才培育项目(gxgnfx2019096);安徽省北斗精准农业信息工程实验室开放基金项目(BDSY2023002);安徽省高校自然科学研究重点项目(KJ2021A1511)。

摘　　要：由于三维复杂图像具有高维度和大规模的特点,且极低比特率下容易导致细节丢失和失真增强,使得三维复杂图像的编码较为复杂。因此,提出基于尺度特征融合的极低比特率三维复杂图像无损压缩方法。利用各向异性扩散与垂直扩散处理三维复杂图像,增强图像边缘信息。采用四叉树算法,建立自适应分块机制,按照图像细节复杂程度划分多个图像块。构建残差网络、反卷积网络结构的残差变换模块,融合图像多尺度特征,输出压缩图像。引入高分辨率累加器和计数器,实现压缩图像无损编码,实现完整的三维复杂图像无损压缩。实验结果表明:应用基于尺度特征融合的新型无损压缩方法后,图像压缩重构结果的信息熵达到了30,实现了压缩图像质量的提升。With its characteristics of high dimension and large scale,3D complex image tends to cause detail loss and distortion enhancement when compressed at very low bit rate,resulting in rather complicated coding.Therefore,a lossless compression method for 3D complex images with very low bit rate based on scale feature fusion was proposed.Anisotropic diffusion and vertical diffusion was used to process 3D complex images and enhance the image edge information.The quadtree algorithm was used to establish an adaptive segmentation mechanism to divide the image into multiple blocks ac‐cording to the complexity of the image details.The residuals transform module of residuals network and deconvolution net‐work structure was set up to fuse multi‐scale features of images and output compressed images.High resolution accumula‐tor and counter were introduced to realize lossless coding of compressed image and implement lossless compression of 3D complex image.The experimental results showed that after the application of the current lossless compression method,which is based on scale feature fusion,the information entropy of image reconstruction results reaches 30,which realizes the improvement of compressed images quality.

关 键 词：尺度特征融合 低比特率 三维图像 无损压缩 深度学习 

分 类 号：TN911.73[电子电信—通信与信息系统]