Abstract: In order to reduce the effects of weak edges and reinforce the shift invariance in sketch-based image re-trieval (SBIＲ)，a multi-scale structure tensor retrieval algorithm on the basis of edge tangent flow field is pro-posed． In this algorithm，edge tangent flow field is used as a substitute for the gradient map of image，and the structure tensor is calculated directly on remarkable strong edges to suppress the effects of weak edges． Besides，structure tensor feature is extracted in the form of multi-scale partition to enhance the shift invariance． Experimental results indicate that the proposed algorithm is superior to the existing structure tensor method because it helps sup-press the effects of weak edges effectively，avoid instable significant edge of image described by image gradient di-rection，enhance the shift invariance of structure，and，thereby，improve the retrieval efficiency．

Key words: image retrieval, sketch, structure tensor, edge tangent flow field