结合形态结构和BP神经网络的盲文字符识别

Braille Recognition Based on the Combination of Character Morphological Structure and BP Neural Network

提出一种基于字符形态结构和BP神经网络的盲文字符识别方法。结合OTSU算法和数学形态学运算实现盲文图像的二值化,通过Canny算子和Hough变换校正盲文图像的旋转畸变;鉴于盲文字符的点位形态结构特征,交替采用形态学膨胀与腐蚀运算处理校正后的盲文图像,确保点位结构的全连接性,通过像素投影法提取盲文字符区域,并结合BP神经网络算法识别盲文字符。结果表明,在40幅测试图像上获得高于95%的盲文字符识别准确率。

Proposes a novel braille recognition method using the combination of character morphological structure and BP neural network.The Otsu al⁃gorithm and mathematical morphology operations were first adopted to obtain the binary segmentation results of the input braille images;then the Canny detector and Hough transform were used to correct the rotation distortion of the threshold braille images.Due to the dotpoint morphological structure of braille characters,the corrected braille images were processed by the alternate use of morphology dilation and erosion operator to guarantee the full connectivity of the dot-points belonging to a unique braille character,and the pixel-wise projec⁃tion was adopted to extract all the potential braille characters from the corrected images.Finally,the extracted braille characters were fed to the BP neural network for braille recognition.Experimental results showed that the proposed method obtained an overall recognition accura⁃cy higher than 95%on 40 test braille images.