A novel method to optimize the polarization state of a polarimetric imaging system is proposed to solve the problem of target/background contrast enhancement in an outdoor environment adaptively. First, the last three elements of the Stokes vector are selected to be the observed object's polarization features, the discriminant projection of which is regarded as the detecting function of our imaging system. Then, the polarization state of the system, which can be seen as a physical classifier, is calculated by training samples with a support vector machine method. Finally, images processed by the system with the designed optimal polarization state become discriminative output directly. By this means, the target/background contrast is enhanced greatly, which results in a more accurate and convenient target discrimination. Experimental results demonstrate that the effectiveness and discriminative ability of the optimal polarization state are credible and stable.