Scene Text Recognition Using Part-Based Tree Structured Models and Linguistic Knowledge

Scene text recognition has nice interests from the computer vision society in recent years. In this paper, a unique scene text-recognition methodology is proposed, which is an integration of structure-guided character detection and linguistic knowledge. In this paper, a part-based tree structure is used to model each and every category of characters to detect and recognize characters at the same time. Since the character models make use of each the native appearance and global structure information, so the detection results are more reliable. For word recognition, combine the detection scores and language model into the posterior probability of character sequence from the Bayesian decision view. The final word-recognition result's obtained by increasing the character sequence posterior probability via Viterbi algorithm. Experimental results on a spread of difficult public datasets demonstrate that the projected methodology achieves progressive performance each for character detection and word recognition.