Bubble size contains important indicative information, which is closely related to flotation production conditions and process indicators. However, due to the different sizes of bubbles in the flotation process and the complexity of the shooting light environment, satisfactory results cannot be obtained from the existing image segmentation methods. In this paper, an improved watershed algorithm based on multiple edge constraints and highlight collegiate positivity is proposed. First, three algorithms are designed to extract and classify foam highlights of the same size, namely, small foam, medium foam and large foam, and special overlap correction and fusion are applied to these three foams. Then, the bubble boundaries are extracted using the Laplace operator, and the segmentation line is constrained with a positive and inverse 45-degree gradient images as multiple edges to ensure the integrity of the segmentation line. Finally, the fused highlight markers are used to deoptimize the external constraint line for watershed segmentation. The tests show that the method is suitable for multiple sizes of fuzzy edges and foam image segmentation. The experimental results show that the accuracy and robustness of the proposed segmentation algorithm are significantly better than other methods, and the proposed method is suitable for foam image segmentation with fuzzy edges and diverse sizes.