Consensus Network Based Hypotheses Combination for Arabic Offline Handwriting Recognition

Abstract

Offline handwriting recognition (OHR) is an extremely challenging task because of many factors including variations in writing style, writing device and material, and noise in the scanning and collection process. Due to the diverse nature of the above challenges, it is highly unlikely that a single recognition technique can address all the characteristics of real-world handwritten documents. Therefore, one must consider designing different systems, each addressing specific challenges in the handwritten corpus, and then combining the hypotheses from these diverse systems. To that end, we present an innovative approach for combining hypotheses from multiple handwriting recognition systems. Our approach is based on generating a consensus network using hypotheses from a diverse set of handwriting recognition systems. Next, we decode the consensus network for producing the best possible hypothesis given an error criterion. Experimental results on an Arabic OHR task show that our combination algorithm outperforms the NIST ROVER technique and results in a 7% relative reduction in the word error rate over the single best OHR system.

Progress in the Raytheon BBN Arabic Offline Handwriting Recognition System

“…SYSTEM COMBINATION Our system combination method is a voting algorithm on top of the output of multiple systems using the consensus network [13]. First, a confusion network is built from n-bests of multiple systems where arcs and nodes of the confusion network represent hypothesized words and transition points, respectively.…”

“…In [13], combined systems differ in image features and training data weights. However, the diversity in this work is limited because complimentary image preprocessing results are not taken into account.…”

“…We combined 4 systems using the consensus networkbased approach [13]. Three systems used preprocessing methods of [27], [28], and [29], respectively (rows 1-3 of TABLE IV) and the last system used both [27] and slant correction (row 4 of TABLE IV).…”

Boost OCR accuracy using iVector based system combination approach

“…To overcome this problem, Fiscus 3 suggested an algorithm called Recognizer Output Voting Error Reduction (ROVER) for speech recognition, where a Word Transition Network (WTN) is generated for word sequence alignment and the best scoring word sequence is extracted by voting with optional word confidence scores. Similarly, Prasad et al 4 proposed a method to combine hypotheses for Arabic OHR by using consensus network (C-Net). To avoid simply using the voting procedure as ROVER, Bertolami et al 5 described a system ensemble method through statistical decision.…”