Adaptive Alpha-Trimmed Average Operator Based on Gaussian Distribution Hypothesis Test for Image Representation

Representation operator is one of the key issues of the content-based retrieval. In this paper, we propose an adaptive alpha-trimmed average operator based on Gaussian distribution hypothesis test for image representation. The adaptive alpha-trimmed average operator extracts the representation by trimming outliers and then estimating the central value of the rest. Since the more samples are used, the more accurate representation we get, the optimal trimming parameter should guarantee to remove the extreme values and at the same time keep useful samples as more as possible. The criterion to distinguish between useful data and extreme noise is derived from Gaussian distribution hypothesis test on the basis of global statistics. Experimental results from standard images show that our proposed scheme outperforms traditional adaptive methods.