Asymmetric Object Recognition with Discriminative Learning and Visual Mirror
Jiabao Wang, Yafei Zhang, Jianjiang Lu, Jifei Chen · Procedia Engineering · 2011
In this paper, a novel approach is proposed for asymmetric object recognition in computer vision. Asymmetric object is very common in the world but often ignored by researchers. It has a visual mirror when we view it from the opposite viewpoint. This property can help us to learn a discriminative model from a large number of labeled images with boxes bounding the interest objects. The novel idea of our approach is that visual mirror of each instance is created to increase the number of training dataset and the position and viewpoint of each instance in the image can be estimated by latent Structural Support Vector Machine (SSVM), which is an important instrument for machine learning in recent years. Experiment proves that our approach has a considerable precision in Caltech101 dataset.