基于自适应下采样和超分重建的图像压缩框架

Image compression framework based on adaptive sub-sampling and super-resolution reconstruction

针对联合图像专家组(JPEG)标准设计了一种基于自适应下采样和超分辨力重建的图像压缩编码框架。在编码器端,为待编码的原始图像设计了多种不同的下采样模式和量化模式,通过率失真优化算法从多种模式中选择最优的下采样模式(DSM)和量化模式(QM),最后待编码图像将在选择的模式下进行下采样和JPEG编码;在解码器端,采用基于卷积神经网络的超分辨力重建算法对解码后的下采样图像进行重建。此外,所提出的框架扩展到JPEG2000压缩标准下同样有效可行。仿真实验结果表明,相比于主流的编解码标准和先进的编解码方法,提出的框架能有效地提升编码图像的率失真性能,并能获得更好的视觉效果。

An image compression coding framework based on adaptive sub-sampling and super-resolution reconstruction is designed for the Joint Photographic Experts Group(JPEG) standard. At the encoder side, a variety of Different Sampling Modes(DSM) and Quantization Modes(QM) are designed for the original image to be encoded. Then, the rate distortion optimization algorithm selects the optimal downsampling and quantization modes from various modes. Finally, the image to be encoded will be sampled and compressed by the standard JPEG compression under the selected optimal mode. In the decoder side, the super-resolution reconstruction algorithm based on convolutional neural network is utilized to reconstruct the decoded sub-sampled image. In addition, the proposed framework is also effective and feasible under the JPEG2000 compression standard. The experimental results show that compared with the mainstream coding and decoding standards and advanced encoding and decoding methods, the framework can effectively improve the rate distortion performance and obtain better visual effects.