基于复合优化的深度玻尔兹曼机的路牌文字图像识别算法

计算机工程与科学

基于复合优化的深度玻尔兹曼机的路牌文字图像识别算法

李文轩,孙季丰

(华南理工大学电子与信息学院,广东广州 510640)

收稿日期:2015-11-20 修回日期:2016-10-17 出版日期:2018-01-25 发布日期:2018-01-25

A traffic signs’Chinese character recognition algorithm

based on mixed optimized deep Boltzmann machine

LI Wen-xuan,SUN Ji-feng

(School of Electronic and Information Engineering,South China University of Technology,Guangzhou 510640,China)

Received:2015-11-20 Revised:2016-10-17 Online:2018-01-25 Published:2018-01-25

摘要/Abstract

摘要：

为提高自然场景中路牌文字图像的识别率，提出一种复合优化的深度玻尔兹曼机文字识别算法。算法以提高目标概率分布的逼近程度为目的，采用两种抽样初始化方法：灰度初始化抽样与二值初始化抽样，构造受限玻尔兹曼机，并由两种初始化方法的受限玻尔兹曼机交叠构成深度玻尔兹曼机。文中提出复合共轭梯度法改进深度玻尔兹曼机的微调算法。实验结果表明，使用文中获取的路牌文字数据，所提算法能够对路牌文字实现较理想的识别效果。与原深度玻尔兹曼机相比，识别率取得有效提高。

关键词: 深度玻尔兹曼机；路牌, 混合初始化；机器学习；文字识别

Abstract:

In order to improve the recognition rate of traffic signs’Chinese characters, we propose a mixed optimized deep Boltzmann machine(MDBM) algorithm to improve the approximation of probability distribution. Two sampling methods (grayscale sampling initialization and binary sampling initialization) are proposed to construct the restricted Boltzmann machines, which are overlapped to form the depth Boltzmann machine. In addition, we propose a fine-tuning algorithm, called complex conjugate gradient method, to improve the fine-tuning part in deep Boltzmann machine. Experiments on traffic signs data show that the recognition rate of the proposed algorithm is better than that of the original deep Boltzmann machine.

Key words: deep Boltzmann machine, traffic signs, mixed initialization, machine learning, Chinese characters recognition

李文轩,孙季丰.

基于复合优化的深度玻尔兹曼机的路牌文字图像识别算法

[J]. 计算机工程与科学.

LI Wen-xuan,SUN Ji-feng.

A traffic signs’Chinese character recognition algorithm

based on mixed optimized deep Boltzmann machine

[J]. Computer Engineering & Science.