This paper first presented a novel and simple method of image segmentation. Experiments illustrates that this method is more effective than some traditional methods, such as maximum-square-error threshold-method and mazimum-entropy threshold-method. In the algorithm of object recognition, we construct object's characters based on four affine monent invariants. Furthermore, we proposed a data-fusion algorithm based on the above characters. The performance of the proposed technique in the recognition experiments of 138 pieces of image, which gained from 18 kinds of objects, is also involved. The experiment result is satisfactory, which the correct-ratio of object recognition is above 85%.