We describe a distributed robust image mosaics system that is robust to outlier and is better in computation efficiency. The system integrates a Web-based user interface, a robust image mosaic, and a distributed computing model. In robust image mosaics, we combine both feature-based and correlation-based approaches for image registration, and then incorporate a skipped mean estimator with Levenberg-Marquardt method for robust nonlinear parameter estimation. The distributed comparing model is used to speed up the computation of the global processing of image mosaics on a network of workstations. Load balancing strategies are proposed to improve the efficiency of parallel computation under uneven load situation. Experimental results show that our system is indeed robust and has significantly speedup in computation.

This paper results from an attempt to unify several different file system design theories. We define a term "partial match pattern" and show that in order to produce file systems optimal with respect to partial match patterns, both the multikey hashing (MKH) method [16] and the multidimensional directory (MDD) method [11] must be in such a form that the number of subdivisions is the same for all domains of keys. We show the conditions for the string homomorphism hashing (SHH) method [15], the MKH method, and the MDD method to be equivalent to one another. We define the so-called Cartesian product files and show that if all records are present, the records in a Cartesian product file form a shortest spanning path in which the Hamming distance between every pair of consecutive records is 1. Thus the SHH method, the MKH method, the MDD method, and the multikey sorting (MKS) method [10] are linked together. Finally, we show that for both partial and best match queries, the file systems exhibit a common characteristic: similar records are grouped together.