基于卷积神经网络的模式分类器  

Building pattern classifiers with convolutional neural networks

作　　者：李葆青[1] 

机构地区：[1]六盘水师范高等专科学校,贵州六盘水553004

出　　处：《大连大学学报》2003年第2期19-23,共5页Journal of Dalian University

摘　　要：研究了使用卷积神经网络((Convolutional Neural Networks,CNNS)构造模式分类器,并用于文本/图像分割和文本检测的可能性.CNNs可以避免显式(直接)特征取样.更为重要的是,CNNs能直接运作于灰度图像,使其应用变得直截了当.对诸如卷积核尺度、网络收敛速度等具体算法实现问题进行了讨论,并给出CNNs在汉字文本/图像分割和文本检测方面的各种实验结果.This paper studies the possibility of buiding pattern clssifiers for text/picture segmentation and text detection problems with convolutional neural networks (CNNs). Using CNN, explicit feature extraction can be avoided. More importantly, CNN can directly operate on grey level images, making its application straightforward. The issues such as kernel size, convergence speed, etc. are discussed. The results of the experiments on Chinese text/ picture segmentation and text detection are presented.

关 键 词：卷积神经网络 模式分类器 灰度图像 人工智能 算法 图像分割 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]