In this dissertation, a novel descriptor is proposed to classify rotated or flipped images/videos. The proposed descriptor is defined as a rotation-and-flipping-robust region binary pattern (RFR). In order to extract RFRs, an input image is converted into a feature map such as luminance or gradient magnitudes. Then, the feature map divided into several rings, and each ring is divided into several sub-regions. Next, mean feature values are computed from each sub-region. From these mean feature values, two types of region binary patterns (RBP) are extracted: intra-RBP and inter-RBP. These RBPs represent horizontal information and vertical information in rings, respectively. Finally, all extracted RBPs are converted into RFR. The proposed RFR has high discrimination since it was based on spatial structure which considers the locations of features in images. In addition, it has the fast feature extraction time and the compact descriptor size.RFR was assessed from two applications which are required to classify rotated or flipped image/videos such as video copy detection and image-based coin recognition. For each application, MUSCLE video copy detection dataset and MUSCLE coin images Seibersdorf dataset were used to compare the-state-of-the-arts, respectively. In the experimental results, RFR showed outperformed results for each application. Especially, RFR could have high recognition and detection rates with smaller descriptor size and shorter feature extraction time. Therefore, the proposed descriptor RFR is very suitable to classify rotated or flipped images/videos.