Project abstract for group papaniko

Sparse Coding for Fast Approximate Nearest Neighbors

These researchers are investigating a new indexing method using sparse coding for fast approximate nearest neighbors (NN) on high-dimensional image data. NN is a fundamental problem in computer science and is used as a core algorithmic component in computer vision applications such as image search, visual surveillance, image registration, etc. Inspired by the recent advances in signal processing and compressive sensing, the idea behind this project is to sparse code each data point using a learned basis dictionary. Indices of the dictionary’s support sets are used to generate one compact identifier for each data point. This generates a small code for each data point, which has fast storage and retrieval using a hash-table mechanism. Typically, most real world datasets worked with in computer vision consist of billions of high dimensional data points, and demand large computational and storage resources for the learning and coding phases. This requires the use of high-performance computing resources.

Return to this PI's main page.