Unknown Object Identification Using Category Visual Words with Rejection Function

In this paper, we introduce an identification method for unknown category objects. Most popular conventional methods in object recognition use Bag of Features (BoF) that represents the image as an appearance frequency histogram of common visual words by quantizing SIFT features. However, this method is unable to identify unknown objects because the common visual words cannot represent the unknown objects well. From this viewpoint, we introduce an unknown object identification method that creates individual category visual words with a rejection function, which can absorb the features of other objects or the background. As a result of object recognition of 10 classes, the proposed method has improved the recognition rate by 8.0 points, compared with the conventional BoF method.