关键词: 小波, 背景标记树（CLT）, 可变权重

Abstract: In this paper，a new multiresolution segmentation algorithm—CLTVWseg，is presented based on contextual information in wavelet domain.In this method，a novel variable weighting parameter is employed to combine the feature random field and the label random field on each scale of wavelet decomposition.These two fields are successively in dominant role to obtain a more accurate segment result by the changing of the weighting parameter during the iterations on each scale.The interscale statistical dependencies between different scales of wavelet decomposition are included in the contextual information，and the segmentation at the finest resolution is used as the final result.Experiments have shown the high efficiency to obtain homogeneous regions and classification boundaries.

Key words: wavelet, Contextual Labeling Tree（CLT）, variable weighting parameter