Vehicle face recognition using weighted visual patches

Distinguishing similar objects is a challenging task; visual patches especially salient or discriminative patches are widely adopted in the state-of-the-art recognition methods to enhance the discovery performance. Considering the fact that different patches have different contributions to the recognition, we develop a fine-grained object recognition algorithm using location and distinction weighted visual patches book, which has two contributions: 1) Location weight is adopted to reduce the influence of non-discriminative patches in the indistinctive area; 2) Between-category differences (DBC) and within-category differences (DWC) are introduced to evaluate the distinction of different patches, which is used to enhance the recognition performance by emphasizing key patches. The paper experimentally demonstrates large improvements over the existing methods for fine-grained as well as position shifted vehicle face recognition.