An improved averaging combination method for image and object recognition

A key development in the design of visual object recognition systems is the combination of multiple features. In recent years, various popular optimization based feature combination methods have been proposed in the literatures. However, those methods obtain tiny performance improvement at the cost of enormous computation consumption. In this paper, we propose an improved averaging combination (IAC) method based on simple averaging combination. Firstly, the discriminative power of features are evaluated by dominant set clustering. Then, these features are ranked and added into the averaging combination one by one in descending order. At last, we obtain the best performance improvement of averaging combination by selecting the most powerful features and removing the weak ones. Experimental results on three challenging datasets demonstrate that our method is order of magnitude faster with competitive and even better results than other sophisticated optimization methods, which can be provided as a better baseline method for feature combination.