Inspired by image salient area detection model, A model for extracting local invariant feature is proposed based on salient measurement of local second moment matrix. The extraction is divided into two stages: The first one is the salient extent estimation of a local feature with analysis of the local image patch’s anisotropy in image scale space. The second one is to direct the feature extraction process according to the extent of local information saliency and to create local salient feature descriptor with more accurate location match and a larger scale span. The experimental results with many real images show that the algorithm is effective and feasible.