A Hybrid Model for Multiple Object Category Detection and Localization.
Dipankar Das, Yoshinori Kobayashi, Yoshinori Kuno · 2009
This paper presents a new hybrid approach to simultaneous detection and localization of multiple object categories using both generative and discriminative models. Our approach consists of first learning the generative model (pLSA) and discriminative model (SVM) using bag of visual words and merging features, respectively. Our merging feature combines spatial shape and appearance of an object. At the same time context graphs are generated from the labeled training datasets. Then, given a new unlabeled test image, a set of promising hypotheses are generated for each object category using pLSA model and bag of visual words representing each object. The discriminative part verifies each hypothesis using SVM classifier with merging features. In the post-processing stage, context information along with the probabilistic output of the SVM classifier is used to improve the overall performance of the system. A combination of features and context information are used to investigate the accuracy of the system. The performance of the proposed framework is evaluated on the various standards (MIT-CSAIL, UIUC, TUD etc.) and the authors ’ own datasets. In experiments we achieved superior results to some state of the art methods over a number of standard datasets. 1