Content-Based Image Retrieval Based on Statistical Properties

This paper proposes a new method, test for equality of covariance andtest for mean vectors, which retrieves different kinds of colours and gray-scale images such as structured, semi-structured, textured, scaled and rotated, and noisy. The proposed method, first, performs preprocessingand then employ the test statistic, such as test for equality of covariance and generalised component test. If the query and target images pass the covariance test, then it is proceeded to test the mean vectors of the two images; otherwise, the test is dropped. If the query and target images pass both tests, then it is concluded that the images are same or similar. Otherwise, they differ. In order to validate the performance of the proposed method, an image database and feature database have been constructed. The images have been collected from various databases to m intain heterogeneity of the images. The proposed method outperforms the existing methods.