A Weighted Edit Distance for Text Correction in Warship Hull Number Recognition

Recognition warship hull number from images is an important application in real world. Generally, the task is transferred as a text recognition issue and tackled by current approaches in street text recognition (STR) field. Post-processing steps are usually introduced for text correcting due to the inevitably error in initial results. Specifically, a domain dictionary consisting of possible results will be collected, and initial results will be compared with items among the dictionary by using edit distance. The item with minimum edit distance is selected as the final recognition result. However, the plain edit distance neglects the shape similarity of characters, resulting in unreasonable matches. In this paper, a weighted edit distance for text correction is proposed. By assigning a low distance for similar characters, the final edit distance can be more accurate for hull number text matching. In addition, a dictionary for warship hull number recognition is constructed for facilitating further research in this field. Experimental results on warship images demonstrate that the proposed approach is effective for warship hull number recognition.