In order to automatically analyze a biological specimen image, a pre-processing step of image feature extraction is often required. During feature extraction, various image features such as pixel intensities, pixel intensity gradients (magnitude and direction), and the like can be extracted from the image. The features can then be used by image analysis tasks such as region segmentation, cell segmentation, scoring, image retrieval, and the like.

However, image feature extraction can be one of the most computationally expensive step in the image analysis pipeline, because high-dimensional features are often required to characterize the complex image contents. In digital pathology, the computational requirements are even higher because of the immense data density of digitized whole slide images. Therefore, in digital pathology and other applications processing high-resolution images of biological specimens, it is desirable to precompute the image features and store them, thereby avoiding multiple redundant computations.