Image Features: Extraction and Categories

We introduce feature-based image classification. Extracted features reduce the data of an image by several orders of magnitude. Ideally, they separate object appearance characteristics from image acquisition artifacts and retain just the former. Artifacts originating from image acquisition and strategies for reducing their impact are presented. Feature extraction is a two-step process. Spatially distributed primary image features represent the image content in a reduced fashion. They are turned into secondary features suitable for classification in feature space. Various examples of features will be discussed. Simple features are textures, edges or corners. They are generated by first filtering the image content followed by a decision criterion to extract features from the result. Advanced features such as histograms of oriented gradients or SIFT extract more descriptive characteristics from the image and are closely related to appearance features of depicted objects. Secondary features are feature vectors that represent each image by the same number of corresponding feature values in a common feature space. Mapping from primary to secondary features will remove some of the spatial information from the features. Different means that retain spatial information to various extents will be presented.