自然场景图像的字符识别方法

目的基于大津算法(Otsu算法)对图像进行分割,利用光学字符识别方法对自然场景图像中的英文字符进行识别。方法首先用分块Otsu算法对图像进行初步的二值化,然后通过对二值化结果的分析,把原始的输入图片分割成单个字符的子图,再对各子图重新用Otsu算法进行二值化,最后对最终得到的二值化结果进行识别,再结合之前得到的每幅图的字符数量信息和词典信息,对识别结果进行修正,得到最终的识别结果。结果在ICDAR2013数据集上测试文中算法,单词正确识别率为46.03%,总编辑距离为474.5。结论文中提出的以Otsu为基础的分块识别算法,能够更好地分割复杂背景图像的背景和文本,同时结合词典信息对识别结果进行了修正,改善了识别效果。

Character Recognition Method in Natural Scene Images

The work aims to segment the image based on the Otsu algorithm and then recognize the English characters in the natural scene images with the method of optical character recognition. First, preliminary binarization of the image was carried with the block Ostu method. Then, the original input image was segmented into sub-graphs of single character after analyzing the binarization results, and all the sub-graphs were binarized again with the Otsu algorithm. Last, the finally obtained binarization results were recognized. Then, the recognition results were modified in combination with the previously obtained information on the number of characters and the dictionary in each image, so as to obtain the final recognition results. The proposed algorithm was tested on the ICDAR2013 dataset. The correct recognition rate of words was 46.03% and the total editing distance was 474.5. The proposed block recognition algorithm based on the Otsu method can better segment the background and the text in complex background images and improve the recognition effect combined with the dictionary information used to modify the recognition results.