The periocular biometric has been shown to be a biometric that is rich in texture and detail. These traits are important for biometric recognition, specifically from the feature extraction module. Traditional texture based feature extraction techniques extract features from an entire biometric image. However, a genetic and evolutionary extraction technique has been created that optimizes the location and dimensions of extraction on biometric images. Optimization occurs on a training set of biometric images that are separated akin to a biometric identification system. Images are placed in a probe set or gallery set and the probe set images are compared to the gallery set images. In this work, a novel variation of the genetic based extraction technique is introduced. The main component of this variation is that the training set is not represented as a traditional identification system, but rather every image in the set is compared with every other set. The results show a superior performance of the variation for periocular identification and verification compared to the traditional genetic approach.