Temporal-Domain Matching of Hand-Drawn Pictorial Queries

In this chapter we discuss the problem of searching a database of hand-drawn documents for a particular hand-drawn query. We show how to break the problem into two parts. First, a matching algorithm must extract pieces of the query and pieces of the database that might be placed in correspondence. Second, the algorithm must examine the pieces in detail to determine how similar they are. We present a general framework that generates potential matches to be examined, and we give two candidate algorithms for the underlying comparison. The rst underlying match algorithm is result of prior work in script matching; the second is a standard dynamic programming-based elastic match. We give experimental data based on 125 database pictures and 125 query pictures to show that our algorithms can eeectively search handwritten data for pictures, even though stroke orders and overall page structure of queries and their intended hits in the database can diier substantially.