摘要
字符识别具有广泛的应用范围,传统的字符识别方法具有局限性,且识别准确率不高。通过对原有算法进行研究,给出了基于骨架提取和HOG特征的字符识别算法,并从初始值、动量项、学习率和网络结构4个方面对BP网络进行优化,以减小局部最小值的可能性。实现了高准确率的低像素图片识别算法。
Character recognition has a wide range of applications,and the traditional character recognition methods have limitations and low recognition accuracy.By studying the original algorithm,a character recognition algorithm based on skeleton extraction and HOG features is given,and the BP network is optimized in four aspects:initial value,momentum term,learning rate and network structure to reduce the possibility of local minima.A high-accuracy low-pixel image recognition algorithm is implemented.
作者
王汝心
马维华
Wang Ruxin;Ma Weihua(Nanjing University of Aeronautics and Astronautics,Nanjing 210016)
出处
《现代计算机》
2021年第24期47-50,共4页
Modern Computer
关键词
BP网络
字符识别
骨架提取
HOG特征
BP network
character recognition
skeleton extraction
HOG features
