A parallel distributed processing algorithm for image feature extraction

We present a new parallel algorithm for image feature extraction. which uses a distance function based on the LZ-complexity of the string representation of the two images. An input image is represented by a feature vector whose components are the distance values between its parts (sub-images) and a set of prototypes. The algorithm is highly scalable and computes these values in parallel. It is implemented on a massively parallel graphics processing unit (GPU) with several thousands of cores which yields a three order of magnitude reduction in time for processing the images. Given a corpus of input images the algorithm produces labeled cases that can be used by any supervised or unsupervised learning algorithm to learn image classification or image clustering. A main advantage is the lack of need for any image processing or image analysis; the user only once defines image-features through a simple basic process of choosing a few small images that serve as prototypes. Results for several image classification problems are presented.