面向感兴趣区域的高性能图像压缩方法

High performance image compression method for region of interest

基于深度学习的图像压缩方法多为有损压缩,而有损压缩通过降低图像质量换取更高的压缩比。为了在码率一定的情况下提高重建图像中感兴趣区域的质量,文中将一种重要性图提取模块嵌入到编码器端,通过提取编码器端最后一层的输出特征来生成重要性图,最终生成掩码用于指导熵编码过程中码率的高效分配。同时,文中将一种解码端增强模块嵌入到解码器输出端,用以预测重建图像中的高频分量,通过增强重建图像中的细节信息来提高重建图像的质量。实验结果表明,以多尺度结构相似性(MS-SSIM)作为评价指标,文中方法优于对比方法,且获得了更好的人眼视觉感知质量。

Most of the image compression methods based on deep learning are lossy compression.Lossy compression reduces image quality in exchange for a higher compression ratio.In order to improve the quality of the region of interest in the reconstructed image under a certain code rate,an importance map extraction module is embedded in the encoder side,and the importance map is generated by extracting the output features of the last layer of the encoder side.The final generated mask is used to guide the efficient allocation of bit rate in the entropy encoding process.At the same time,a decoder-side enhancement module is embedded in the decoder output to predict the high-frequency components in the reconstructed image,and improve the quality of the reconstructed image by enhancing the detail information in the reconstructed image.The experimental results show that with multi-scale structural similarity(MS-SSIM)as the evaluation index,the method in this paper is better than the comparison method and obtains better visual perception quality in human eyes.