Abstract: Feature parameters of froth image in flotation process are closely associated with the industrial situations. This study has presented an improved watershed algorithm based on adaptive morphological operation, considering lack of background and various non-uniform froths conglutinating together. Firstly, Two-Dimensional Histogram Minimum Cross-Entropy (GLCM) is used to estimate the froth distribution, which is divided into three regions: large, middle and small. Secondly, the image is enhanced by multi-scale Retinex algorithm, due to the effects imposed by light and noises. Then, the optimal structural element is used to perform morphological filtering according to regions types and extract mark points using adaptive area morphology reconstruction. Lastly, results are obtained through watershed algorithm. Simulation indicated that proposed algorithm is more robust in segmenting unevenly-distributed froth image, overcoming over-segmentation and under-segmentation effectively.