基于自适应高斯混合模型的JPEG压缩图像去块效应算法

Image Deblocking based on Adaptive Gaussian Mixture Model

基于分块离散余弦变换的JPEG图像压缩算法,在低码率时解码图像会产生明显的块效应,严重降低了图像质量。针对JPEG压缩图像产生的块效应,提出了一种利用高斯混合模型自适应学习图像先验的去块效应算法。该算法利用初始去块效应图像将外部图像库训练得到的高斯混合模型映射到针对特定图像的高斯混合模型,得到自适应的高斯混合模型,然后结合加权的稀疏表示方法,利用训练得到的自适应高斯混合模型用于图像去块效应。实验结果表明,提出的去块效应算法能较好地去除块效应,且优于一些图像去块效应和图像去噪算法。

The JPEG image compression algorithm based on the block discrete cosine transform, when at low bit rate, would usually produce obvious block effect, and this may seriously degrade the image quality. In view of the block effect produced by the JPEG compressed image, the deblocking algorithm which uses the Gauss hybrid model to learn the image prior is proposed. In the algorithm, the Gauss mixture model trained by the external image library is mapped, by using the initial deblocking effect, to the Gauss mixture model for the specific image, and an adaptive Gauss mixture model is acquired Then combined with the weighted sparse representation, the acquired adaptive Gauss hybrid model is used for the image deblocking effect. The experimental results show that the proposed deblocking algorithm could fairly remove the block effect, and is superior to some image deblocking and image denoising algorithms.