论文部分内容阅读
在传统的图像压缩编码中,只对幅度值较大的变换系数进行编码,而舍弃幅度值较小的变换系数(以下称“系数”皆指变换后的系数),以实现比较高的压缩率。幅度值较小的系数对应图像的细节,也就是高频纹理信息。那么码率越低,需要舍弃的系数就越多,细节失真就越大。人的视觉对一些图像的纹理失真是比较敏感的,比如人脸。文中在JPEG 2000的基础上,用基于树结构的纹理建模方法(TM BT)在基本码流中加入了纹理信息,在不大幅度增加码率和计算量的情况下,较好地保持了图像细节,改善了视觉效果。
In traditional image compression coding, only the transform coefficient with larger amplitude value is encoded, and the transform coefficient with smaller amplitude value (hereinafter referred to as “coefficient” refers to the transformed coefficient) is discarded to achieve relatively high Compression ratio. The smaller the magnitude of the coefficient corresponding to the image details, that is, high-frequency texture information. Then the lower the code rate, the more coefficients that need to be discarded and the greater the detail distortion. Human vision is more sensitive to textural distortions of some images, such as faces. Based on JPEG2000, texture information is added to the basic stream by using the tree structure-based texture modeling method (TM BT), which keeps the bit rate and computational complexity unchanged Image details to improve visual effects.