IMPLEMENTATION OF SPARSE CODING FOR EVENT TRACKING AND IMAGE RETRIEVAL
Abstract
Comparing regions of images is a fundamental task in both similarity based object tracking as well as retrieval of images from image datasets, where an exemplar image is used as the query. In this thesis, we focus on the task of creating a method of comparison for images produced by NASA’s Solar Dynamic Observatory mission. This mission has been in operation for several years and produces almost 700 Gigabytes of data per day from the Atmospheric Imaging Assembly instrument alone. This has created a massive repository of high-quality solar images to analyze and categorize. To this end, we are concerned with the creation of image region descriptors that are selective enough to differentiate between highly similar images yet compact enough to be compared in an efficient manner, while also being indexable with current indexing technology. We produce such descriptors by pooling sparse coding vectors produced by spanning learned basis dictionaries. Various pooled vectors are used to describe regions of images in event tracking, entire image descriptors for image comparison in content based image retrieval, and as region descriptors to be used in a content based image retrieval system on the SDO AIA image pipeline.
IMPLEMENTATION OF SPARSE CODING FOR EVENT TRACKING AND IMAGE RETRIEVAL. GET MORE COMPUTER SCIENCE PROJECT TOPICS AND MATERIALS
