In order to identify maize purity rapidly and efficiently, the image processing technology and clustering algorithm were studied according to the maize seed and characteristics of the seed images. An improved DBSCAN on the basis of farthest first traversal algorithm (FFT) adapting to maize seeds purity identification was proposed. The color features parameters of the RGB, HIS and Lab color models of maize crown core area were extracted. H, S and B were selected to be the effective characteristic vector. The abnormal points of different density characteristic vector points were separated by FFT. Then clustering results were combined after local density cluster by DBSCAN. Experiment results showed that the method played a great role in improving the accuracy of maize purity identification. 