3D Haar-like elliptical features for object classification in microscopy

Object detection and classification are key tasks in computer vision that can facilitate high-throughput image analysis of microscopy data. We present a set of local image descriptors for three-dimensional (3D) microscopy datasets inspired by the well-known Haar wavelet framework. We add orientation, illumination and scale information by assuming that the neighborhood surrounding points of interests in the image can be described with ellipsoids, and we increase discriminative power by incorporating edge and shape information into the features. The calculation of the local image descriptors is implemented in a Graphics Processing Unit (GPU) in order to reduce computation time to 1 millisecond per object of interest. We present results for cell division detection in 3D time-lapse fluorescence microscopy with 97.6% accuracy.