Abstract: Due to parameters is needed to be set depending on the issue of human goal-oriented to filter matting algorithms， a classifier was proposed based on component detection called LSVM (latent support vector machine). The parameters of guided radius could be automatically set so as to optimize algorithm. Firstly， for determining the radius， a potential idea of LSVM was applied which using the known database templates to train the input samples of the target image， then the sample set of the guided image and the binary image were divided into blocks with different radius which could automatically generate the reasonable parameters and optimize algorithm. At last， guided filter was used to advance matting and the intelligent and flexible of the matting algorithm was achiered.

Key words: guided filter, image matting, LSVM(latent support verctor machine), target image, training sample set