This　paper　propose　an　image　segmentation　algorithm　for　mineral　froth　images,　based　on　multi-scale　analysis,　valley　edge　detection,　edge　thinning,　end　point　detection　and　linking,　and　region　merge　on　shape.　The　algorithm　was　tested　for　a　number　of　different　froth　images　and　was　compared　to　the　watershed,　similarity　based　and　Canny　operator　based　algorithms.　The　testing　results　show　that　the　similarity　or　edge　based　algorithms　cannot　segment　froth　images　well,　and　the　watershed　algorithm　often　causes　over-segmentation,　especially　for　the　images　of　relative　large　bubbles.　The　accuracy　and　processing　speed　of　the　studied　algorithm　are　better　than　that　of　the　watershed　algorithm.　The　test　results　show　that　the　algorithm　performs　satisfactorily.