Abstract
We presented a new algorithm of underwater bubble recognition, which employs background modeling, image segmentation and pattern recognition. After obtaining underwater bubble images, we can separate single bubble from it manually and construct the database. Having computed Hu moment of samples for training and test, we can get the threshold and store. Then inputting the other images of sample, we can discriminate the bubbles by the threshold. Next, by calculating their gradients from the center to the edge in different vectors, we detect their gradients and estimate whether they keep consistent. Finally, the accuracy of bubble recognition is up to 94%. It is concluded that this algorithm not only recognizes bubbles to filter suspensions in irregular shapes by Hu moment, but also adds gradients calculation for further estimation, so it is high in precision.