一种具有旋转不变性的压缩域纹理图像分类方法

TEXTURE-IMAGE CLASSIFICATION WITH ROTATION-INVARIANCE IN COMPRESSED-DOMAIN

该文提出了一种具有旋转不变性的、基于DCT压缩域的纹理图像分类方法。此方法主要是在DCT压缩域中合理组织频率系数,使其具有与小波分解相类似的图像多分辨率结构,由此设计了抗旋转能力强的纹理图像分类算法。这些操作直接在游程长编码(RLE)形成的码流上进行,处理的数据量较少,处理速度快。实验结果表明,这种分类方法具有较高的准确性。

A novel method of texture-image classification with rotation-invariauce in DCT compressed-domain is proposed. By reordered the DCT coefficients for multi-resolution, the textnral features are extracted in DCT transformed-domain and these textural features can be used to classify texture-images in RLE bit-stream directly with rotation-invariancc. The experimental results demonstrate that texture-images can be classified at higher correct classification rate.