基于图像纹理的混合图像压缩算法

Hybrid Image Compression Algorithm Based on Image Texture

摘要: 针对经典图像压缩编码中跨纹理边缘图像块编码性能差、存在非稀疏表示等问题，提出一种基于图像纹理的混合图像压缩编码方案。在图傅里叶变换中，利用图像块具有不同纹理特征的特点对图拉普拉斯矩阵进行构造并实现图像块纹理特征的自适应选择，以提高图像块的压缩编码性能；再通过率失真优化实现离散余弦变换和图傅里叶变换的最优变换选择。实验结果表明，在相同比特率下，混合图像压缩算法的峰值信噪比比单纯的离散余弦变换平均高出1.08 dB，并且在相同的量化步长下具有更好的图像质量。

Abstract: A hybrid image compression scheme based on image texture is proposed to address the problems of poor coding performance of image blocks across texture edges and the existence of non-sparse representation in classical image compression coding. Firstly, the image blocks with different texture features are utilized in the graph Fourier transform to construct the graph La-place matrix and realize the adaptive selection of texture features of image blocks to improve the compression coding performance of image blocks. Secondly, the optimal transform selection is achieved by rate-distortion optimization in discrete cosine transform and graph Fourier trans-form. Experimental results show that the hybrid image compression algorithm outperforms a simple discrete cosine transform by an average of 1.08 dB in peak signal-to-noise ratio at the same bit rate, and it has better image quality at the same quantization step.