In order to analyze the image texture effectively, the new rotation invariant and multiresolution texture descriptors are proposed based on Local Walsh Spectrum (LWS). Firstly, the Local Binary Sequence (LBS) of each pixel is obtained by comparing its gray-scale with neighboring points and the power spectrum of Discrete Walsh Transform (DWT) of the LBS is calculated. Then, the spectrum’s value in the power spectrum is expressed in characteristic histogram to describe the texture feature. Finally, based on the sequency characteristic of LBS, the Two-family Sequency LWS (TSLWS) is proposed to reveal the relationship between LWS and Local Binary Pattern (LBP).Because of the circular-shift-invariant of DWT power spectrum, the proposed texture descriptors show prior rotation invariance. Experimental results indicate that the texture classification precisions of LWS are better than those of the Gray Level Co-occurrence Matrix (GLCM) method and Gabor filter bank method. Furthermore,as compared with the LBP,the texture classification precision of the LWS is 3% higher than that of LBP in the same local neighborhood and the segmentation in inaccuracies of the LWS are 11% and 23% respectively less than those of LBP for two rotated mosaic texture images,which proves that the proposed method has better abilities of texture discrimination and rotation invariance.孙慧贤|Saber_sun@163.com