激光与光电子学进展, 2019, 56 (8): 081007, 网络出版: 2019-07-26
基于卷积神经网络的棋子定位和识别方法 下载: 1330次
Methods for Location and Recognition of Chess Pieces Based on Convolutional Neural Network
图像处理 卷积神经网络 二值图像滤波 深度学习 棋子定位 棋子识别 image processing convolutional neural network binary image filtering deep learning location of chess pieces recognition of chess pieces
摘要
中国象棋棋子定位采用的传统图像处理方法,复杂度高;识别棋子采用的传统文字识别方法,泛化性较差、精确度较低。提出一种基于棋子颜色特征的分割方法和改进的二值图像滤波算法,实现了棋子的快速定位,不需要二次修正位置;提出一种基于卷积神经网络的棋子识别方法,该方法可以应用于不同字体的棋子识别,在更换棋子的情况下,依然可以快速、准确地识别棋子。实验结果表明,该方法的定位误差为0.51 mm,平均定位时间0.212 s,对4类字体的平均棋子识别准确率为98.59%左右,证实了该方法的有效性和实用性。
Abstract
The traditional image processing algorithms used for the location of Chinese chess pieces have high complexity and the traditional character recognition methods used for the recognition of chess pieces have low generalization and accuracy. A segmentation method based on chess piece color features and an improved binary image filtering algorithm are proposed to achieve the fast location of chess pieces, and the second correction of positions is not needed. A recognition method of chess pieces based on a convolutional neural network is proposed, which can be used for the recognition of chess pieces with different fonts. In the case of chess piece replacement, this method can still recognize chess pieces quickly and accurately. The experimental results show that as for the proposed method, the location error is 0.51 mm, the average location time is 0.212 s, and the average recognition accuracy of chess pieces with four types of fonts is about 98.59%. The effectiveness and practicability of this method are confirmed.
韩燮, 赵融, 孙福盛. 基于卷积神经网络的棋子定位和识别方法[J]. 激光与光电子学进展, 2019, 56(8): 081007. Xie Han, Rong Zhao, Fusheng Sun. Methods for Location and Recognition of Chess Pieces Based on Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(8): 081007.