Statistical Texture Feature Extraction Based on Wavelet Compressed Domain
Shen Lan-sun · Dianzi xuebao · 2003
Discrete wavelet transform has often been used for multi-scale texture characterization through the analysis of spatial-frequency content.Most previous methods make no account of the correlation of coefficients in adjacent subbands.However,in intuition,the coefficients in adjacent subbands are highly correlated.A novel method for texture feature extraction is now proposed based on the statistics relationships of wavelet coefficients at adjacent scale subbands with the same orientation.In addition,taking into account the fact that most of current images are stored and transmitted in the compressed format,we try to make the proposed method compatible with the new generation image compression standard-JPEG2000.Therefore,texture classification can be performed directly on the compressed DWT domain(just entropy decoding needed).Experimental results show that the proposed scheme has outperformed the previous methods,and the best performance is achieved by combining cross-subband relationship and traditional subband energy.