Conventional KNN (K-nearest Neighbor) algorithm requires large computational consumption when training samples are large. To address this concern, a novel immune-based data reduction approach was proposed and applied to solve color recognition of license plates. Antigenic determinant, immune cells, etc, were defined and computational method of affinity was given. In the process of multi-species parallel learning, clone selection and mutation, immune tolerance, immune memory and other mechanisms were employed to implement data reduction on each training antigen set. Then the reduced representative detectors were used to perform color recognition of license plates during the second immune response stage, integrated with KNN approach. Experiments were conducted on two data sets, compared with three data reduction approaches based on histograms,and results showed that the proposed algorithm can perform data reduction task effectively, with the reduction rates of 98.87% and 95.48%, respectively,and the overall accuracies of 97.45% and 94.73%,
