Abstract: Considering the redundancy of froth image characters that reflect the status of the flotation process,an unsupervised reduction method is proposed for selecting froth image characters for flotation. First,a sensitivity index is defined,and sensitive characters are set. These characters,which are strongly related to the status of the flotation process,are obtained through a reduction method based on the sensitivity index. Then,for solving the correlation problem that exists in sensitive image characters,a reduction method for sets of sensitive image characters is developed using rough set attribute importance. Finally,application to the flotation of gold and antimony and tests based on industrial local data demonstrate the method's effectiveness,which improves machine-vision-based process monitoring of froth flotation.