The verification of vehicle License Plates (LPs) has not been given much importance in existing LP recognition systems as only a handful of methods deal with this problem explicitly. For an efficient system, it is imperative that a detected LP is validated first before the recognition of characters on it. Majority of the existing methods make use of geometrical constraints for the elimination of false LP regions which is not an effective way as multinational LPs have variable geometrical attributes and diversity in styles. To overcome these limitations, in this paper, we evaluate three kinds of representative local descriptors (SURF, HOG and LBP) and their combinations along with AIexNet CNN for the classification of LP and non-LP regions to provide a unique solution for the validation of multinational LPs. Experiments on 13490 LP and non-LP images show that the HOG feature individually gives the best recognition rate of 96.94% while considering collectively, best of 98.35% is achieved for SURF+HOG; whereas, the fine-tuned AIexNet outperform all others in terms of recognition accuracy of 99.27% but requires extensive processing. Furthermore, the proposed model is incorporated in one of the existing LP detection methods to demonstrate improved performance.