Keyword Spotting Scores Fusion based on Fuzzy Integral and Curvelet Descriptor

摘要

This paper deals with a new matching scheme for keyword spotting that combines advantages of two set of different descriptors using the Discrete Choquet Fuzzy Integral (DCFI) combiner and the Dynamic Time Warping (DTW) algorithm. In fact, DTW computes for each descriptor the score of matching between two word images. However, the idea of FI technique is that the final matching decision can be significantly increased by giving the combiner a possibility to bias the different DTW output scores based on a priori knowledge about the worthiness degree of each descriptor. Certainly, a complementary information can be derived from two matching system with two different inputs. The experiments conducted on two different document images show that the fusion of statistical and Curvelet descriptors provides a higher spotting precision in comparison with other descriptors.